update server

app.py

@@ -6,7 +6,7 @@ from serve.gradio_web_i2s import *
 from serve.leaderboard import build_leaderboard_tab
 from model.model_manager import ModelManager
 from pathlib import Path
-from serve.constants import SERVER_PORT, ROOT_PATH, ELO_RESULTS_DIR
+from constants import SERVER_PORT, ROOT_PATH, ELO_RESULTS_DIR
 
 
 def build_combine_demo(models, elo_results_file, leaderboard_table_file):

serve/constants.py → constants.py

@@ -1,8 +1,12 @@
 import os
+from pathlib import Path
+
+os.chdir(Path.cwd())
 
 LOGDIR = os.getenv("LOGDIR", "./3DGen-Arena-logs/vote_log")
 IMAGE_DIR = os.getenv("IMAGE_DIR", f"{LOGDIR}/images")
 OFFLINE_DIR = "./offline"
+OFFLINE_GIF_DIR = os.path.join(OFFLINE_DIR, "gifs")
 
 SERVER_PORT = os.getenv("SERVER_PORT", 7860)
 ROOT_PATH = os.getenv("ROOT_PATH", None)
@@ -17,4 +21,8 @@ SAVE_IMAGE = "save_image"
 SAVE_LOG = "save_log"
 
 NUM_SIDES = 2
-TEXT_PROMPT_PATH = "offline/prompts.json"
+TEXT_PROMPT_PATH = "offline/prompts_110.json"
+IMAGE_PROMPT_PATH = "offline/image_urls.txt"
+
+MAX_ATTEMPTS = 5
+REPLICATE_API_TOKEN = os.getenv("REPLICATE_API_TOKEN", "r8_0BaoQW0G8nWFXY8YWBCCUDurANxCtY72rarv9")

model/client.py

@@ -0,0 +1,22 @@
+import os
+from gradio_client import Client
+from constants import MAX_ATTEMPTS
+
+class GaussianToMeshClient:
+    def __init__(self):
+        self.client = Client("https://dylanebert-splat-to-mesh.hf.space/", upload_files=True, download_files=True)
+    
+    def run(self, shape):
+        attempt = 1
+        mesh = None
+        while attempt <= MAX_ATTEMPTS:
+            try:
+                mesh = self.to_mesh_client.predict(shape, api_name="/run")
+                break
+            except Exception as e:
+                print(f"Attempt to convert Gaussian to Mesh Failed {attempt}/{MAX_ATTEMPTS}: {e}")
+                attempt += 1
+                # time.sleep(1)
+        return mesh
+    
+Gau2Mesh_client = GaussianToMeshClient()

model/model_config.py

@@ -38,14 +38,44 @@ register_model_config(
     online_model=False
 )
 
+# register_model_config(
+#     model_name="mvdream",
+#     i2s_model=False,
+#     online_model=False
+# )
+
+# register_model_config(
+#     model_name="prolificdreamer",
+#     i2s_model=False,
+#     online_model=False
+# )
+
 register_model_config(
-    model_name="mvdream",
-    i2s_model=False,
+    model_name="dreamgaussian",
+    i2s_model=True,
     online_model=False
 )
 
+# register_model_config(
+#     model_name="wonder3d",
+#     i2s_model=True,
+#     online_model=False
+# )
+
 register_model_config(
-    model_name="prolificdreamer",
-    i2s_model=False,
+    model_name="lgm",
+    i2s_model=True,
+    online_model=False
+)
+
+register_model_config(
+    model_name="openlrm",
+    i2s_model=True,
+    online_model=False
+)
+
+register_model_config(
+    model_name="triplane-gaussian",
+    i2s_model=True,
     online_model=False
 )

model/model_manager.py

@@ -61,47 +61,66 @@ class ModelManager:
             return 
 
     # @spaces.GPU(duration=120)
-    def inference(self, prompt, model_name):
+    def inference(self, 
+                  prompt, model_name,
+                  offline=False, offline_idx=None):
+        result = None
         worker = self.models_worker[model_name]
-        result = worker.inference(prompt=prompt)
+
+        if offline:
+            result = worker.load_offline(offline, offline_idx)
+        if not offline or result == None:
+            if worker.check_online():
+                result = worker.inference(prompt)
         return result
     
-    def render(self, prompt, model_name):
+    def render(self, shape, model_name):
         worker = self.models_worker[model_name]
-        result = worker.render(prompt=prompt)
+        result = worker.render(shape)
         return result
     
-    def inference_parallel(self, prompt, model_A, model_B):
+    def inference_parallel(self, 
+                           prompt, model_A, model_B, 
+                           offline=False, offline_idx=None):
         results = []
         model_names = [model_A, model_B]
         with concurrent.futures.ThreadPoolExecutor() as executor:
-            future_to_result = {executor.submit(self.inference, prompt, model): model 
+            future_to_result = {executor.submit(self.inference, prompt, model, offline, offline_idx): model 
                                 for model in model_names}
             for future in concurrent.futures.as_completed(future_to_result):
                 result = future.result()
                 results.append(result)
         return results[0], results[1]
 
-    def inference_parallel_anony(self, prompt, model_A, model_B, i2s_model):
+    def inference_parallel_anony(self, 
+                                 prompt, model_A, model_B, 
+                                 i2s_model: bool, offline: bool =False, offline_idx: int =None):
         if model_A == model_B == "":
-            model_A, model_B = random.sample(self.get_models(i2s_model=i2s_model, online_model=True), 2)
+            if offline and i2s_model:
+                model_A, model_B = random.sample(self.get_i2s_models(), 2)
+            elif offline and not i2s_model:
+                model_A, model_B = random.sample(self.get_t2s_models(), 2)
+            else:
+                model_A, model_B = random.sample(self.get_models(i2s_model=i2s_model, online_model=True), 2)
         model_names = [model_A, model_B]
+
         results = []
         with concurrent.futures.ThreadPoolExecutor() as executor:
-            future_to_result = {executor.submit(self.inference, prompt, model): model 
+            future_to_result = {executor.submit(self.inference, prompt, model, offline, offline_idx): model 
                                 for model in model_names}
             for future in concurrent.futures.as_completed(future_to_result):
                 result = future.result()
                 results.append(result)
-        return results[0], results[1]
+        return results[0], results[1], model_A, model_B
     
     
-    def render_parallel(self, prompt, model_A, model_B):
+    def render_parallel(self, shape_A, model_A, shape_B, model_B):
         results = []
         model_names = [model_A, model_B]
+        shapes = [shape_A, shape_B]
         with concurrent.futures.ThreadPoolExecutor() as executor:
-            future_to_result = {executor.submit(self.render, prompt, model): model 
-                                for model in model_names}
+            future_to_result = {executor.submit(self.render, shape, model): model 
+                                for model, shape in zip(model_names, shapes)}
             for future in concurrent.futures.as_completed(future_to_result):
                 result = future.result()
                 results.append(result)

model/model_registry.py

@@ -52,13 +52,6 @@ register_model_info(
     "Text-to-3D using 2D Diffusion and SDS Loss",
 )
 
-register_model_info(
-    ["dreamgaussian"],
-    "DreamGaussian",
-    "https://github.com/dreamgaussian/dreamgaussian",
-    "Generative Gaussian Splatting for Efficient 3D Content Creation",
-)
-
 register_model_info(
     ["fantasia3d"],
     "Fantasia3D",
@@ -66,6 +59,13 @@ register_model_info(
     "Disentangling Geometry and Appearance for High-quality Text-to-3D Content Creation",
 )
 
+register_model_info(
+    ["instant3d"],
+    "Instant3D",
+    "https://jiahao.ai/instant3d/",
+    "Fast Text-to-3D with Sparse-View Generation and Large Reconstruction Model",
+)
+
 register_model_info(
     ["latent-nerf"],
     "Latent-NeRF",
@@ -87,6 +87,14 @@ register_model_info(
     "Disentangling 2D and Geometric Priors for High-Fidelity and Consistent 3D Generation",
 )
 
+
+register_model_info(
+    ["lucid-dreamer"],
+    "LucidDreamer",
+    "https://github.com/EnVision-Research/LucidDreamer",
+    "Towards High-Fidelity Text-to-3D Generation via Interval Score Matching",
+)
+
 register_model_info(
     ["mvdream"],
     "MVDream",
@@ -94,6 +102,20 @@ register_model_info(
     "Multi-view Diffusion for 3D Generation",
 )
 
+register_model_info(
+    ["point-e"],
+    "Point·E",
+    "https://github.com/openai/point-e",
+    "A System for Generating 3D Point Clouds from Complex Prompts",
+)
+
+register_model_info(
+    ["shap-e"],
+    "Shap-E",
+    "https://github.com/openai/shap-e",
+    "Generating Conditional 3D Implicit Functions",
+)
+
 register_model_info(
     ["prolificdreamer"],
     "ProlificDreamer",
@@ -108,6 +130,30 @@ register_model_info(
     "Generating Multiview-consistent Images from a Single-view Image",
 )
 
+register_model_info(
+    ["sjc"],
+    "Score Jacobian Chaining",
+    "https://pals.ttic.edu/p/score-jacobian-chaining",
+    "Lifting Pretrained 2D Diffusion Models for 3D Generation",
+)
+
+# register_model_info(
+#     [],
+#     "",
+#     "",
+#     "",
+# )
+
+
+## regist image-to-shape generation models
+register_model_info(
+    ["dreamgaussian"],
+    "DreamGaussian",
+    "https://github.com/dreamgaussian/dreamgaussian",
+    "Generative Gaussian Splatting for Efficient 3D Content Creation",
+)
+
+
 register_model_info(
     ["wonder3d"],
     "Wonder3D",
@@ -122,17 +168,6 @@ register_model_info(
     "Hierarchical 3d generation with bootstrapped diffusion prior",
 )
 
-
-# register_model_info(
-#     [],
-#     "",
-#     "",
-#     "",
-# )
-
-
-# regist image edition models
-
 register_model_info(
     ["zero123"],
     "Zero-1-to-3",
@@ -160,14 +195,6 @@ register_model_info(
     "https://github.com/bytedance/ImageDream",
     "Image-Prompt Multi-view Diffusion for 3D Generation",
 )
-
-register_model_info(
-    ["lucid-dreamer"],
-    "LucidDreamer",
-    "https://github.com/EnVision-Research/LucidDreamer",
-    "Towards High-Fidelity Text-to-3D Generation via Interval Score Matching",
-)
-
 register_model_info(
     ["make-it-3d"],
     "Make-It-3D",
@@ -210,6 +237,27 @@ register_model_info(
     "Large Multi-View Gaussian Model for High-Resolution 3D Content Creation",
 )
 
+register_model_info(
+    ["gsgen"],
+    "GSGEN",
+    "https://github.com/gsgen3d/gsgen",
+    "Text-to-3D using Gaussian Splatting",
+)
+
+register_model_info(
+    ["openlrm"],
+    "OpenLRM",
+    "https://github.com/3DTopia/OpenLRM",
+    "Open-Source Large Reconstruction Models",
+)
+
+register_model_info(
+    ["hifa"],
+    "HiFA",
+    "https://github.com/JunzheJosephZhu/HiFA",
+    "High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance",
+)
+
 # register_model_info(
 #     [],
 #     "",

model/model_worker.py

@@ -1,40 +1,55 @@
 import os
+import json
 import time
+import kiui
 from typing import List
 import replicate
+import subprocess
 
-os.environ("REPLICATE_API_TOKEN", "r8_0BaoQW0G8nWFXY8YWBCCUDurANxCtY72rarv9")
+import sys
+sys.path.append("..")
+
+from gradio_client import Client
+from constants import OFFLINE_GIF_DIR, MAX_ATTEMPTS, REPLICATE_API_TOKEN
+from .client import Gau2Mesh_client
+
+# os.environ["REPLICATE_API_TOKEN"] = REPLICATE_API_TOKEN
 
 class BaseModelWorker:
     def __init__(self,
                  model_name: str,
                  i2s_model: bool, 
                  online_model: bool,
-                 model_path: str = None,
+                 model_api: str = None
                  ):
         self.model_name = model_name
         self.i2s_model = i2s_model
         self.online_model = online_model
-        self.model_path = model_path
-        self.model = None
+        self.model_api = model_api
+        self.urls_json = None
         
-        if self.online_model:
-            assert not self.model_path, f"Please give model_path of {model_name}"
-            self.model = self.load_model()
+        urls_json_path = os.path.join(OFFLINE_GIF_DIR, f"{model_name}.json")
+        if os.path.exists(urls_json_path):
+            with open(urls_json_path, 'r') as f:
+                self.urls_json = json.load(f)
 
     def check_online(self) -> bool:
         if self.online_model and not self.model:
             return True
         else:
             return False
-        
-    def load_model(self):
-        pass
     
+    def load_offline(self, offline: bool, offline_idx):
+        ## offline 
+        if offline and str(offline_idx) in self.urls_json.keys():
+            return self.urls_json[str(offline_idx)]
+        else:
+            return None
+
     def inference(self, prompt):
         pass
 
-    def render(self, shape):
+    def render(self, shape, rgb_on=True, normal_on=True):
         pass
 
 class HuggingfaceApiWorker(BaseModelWorker):
@@ -44,25 +59,25 @@ class HuggingfaceApiWorker(BaseModelWorker):
             i2s_model: bool, 
             online_model: bool,
             model_api: str,
-            model_path: str = None,
     ):
         super().__init__(
             model_name,
             i2s_model, 
             online_model,
-            model_path,
+            model_api,
         )
-        self.model_api = model_api
 
 class PointE_Worker(BaseModelWorker):
     def __init__(self, 
                  model_name: str, 
                  i2s_model: bool, 
                  online_model: bool, 
-                 model_api: str,
-                 model_path: str = None):
-        super().__init__(model_name, i2s_model, online_model, model_path)
-        self.model_api = model_api
+                 model_api: str):
+        super().__init__(model_name, i2s_model, online_model, model_api)
+
+class TriplaneGaussian(BaseModelWorker):
+    def __init__(self, model_name: str, i2s_model: bool, online_model: bool, model_api: str = None):
+        super().__init__(model_name, i2s_model, online_model, model_api)
 
 
 class LGM_Worker(BaseModelWorker):
@@ -70,26 +85,69 @@ class LGM_Worker(BaseModelWorker):
                  model_name: str, 
                  i2s_model: bool, 
                  online_model: bool, 
-                 model_path: str = "camenduru/lgm-ply-to-glb:eb217314ab0d025370df16b8c9127f9ac1a0e4b3ffbff6b323d598d3c814d258"):
-        super().__init__(model_name, i2s_model, online_model, model_path)
+                 model_api: str = "camenduru/lgm:d2870893aa115773465a823fe70fd446673604189843f39a99642dd9171e05e2",
+    ):
+        super().__init__(model_name, i2s_model, online_model, model_api)
+        self.model_client = replicate.Client(api_token=REPLICATE_API_TOKEN)
     
     def inference(self, image):
-        output = replicate.run(
-            self.model_path,
-            input={"ply_file_url": image}
+        
+        output = self.model_client.run(
+            self.model_api,
+            input={"input_image": image}
         )
-        #=> .glb file url: "https://replicate.delivery/pbxt/r4iOSfk7cv2wACJL539ACB4E...
-        return output
+        #=> .mp4 .ply
+        return output[1]
+
+    def render(self, shape):
+        mesh = Gau2Mesh_client.run(shape)
+
+        path_normal = ""
+        cmd_normal = f"python -m ..kiuikit.kiui.render {mesh} --save {path_normal} \
+            --wogui --H 512 --W 512 --radius 3 --elevation 0 --num_azimuth 40 --front_dir='+z' --mode normal"
+        subprocess.run(cmd_normal, shell=True, check=True)
+
+        path_rgb = ""
+        cmd_rgb = f"python -m ..kiuikit.kiui.render {mesh} --save {path_rgb} \
+            --wogui --H 512 --W 512 --radius 3 --elevation 0 --num_azimuth 40 --front_dir='+z' --mode rgb"
+        subprocess.run(cmd_rgb, shell=True, check=True)
+        
+        return path_normal, path_rgb
+
+class V3D_Worker(BaseModelWorker):
+    def __init__(self, 
+                model_name: str, 
+                i2s_model: bool, 
+                online_model: bool, 
+                model_api: str = None):
+        super().__init__(model_name, i2s_model, online_model, model_api)
+
+
+# model = 'LGM'
+# # model = 'TriplaneGaussian'
+# folder = 'glbs_full'
+# form = 'glb'
+# pose = '+z'
+
+# pair = ('OpenLRM', 'meshes', 'obj', '-y')
+# pair = ('TriplaneGaussian', 'glbs_full', 'glb', '-y')
+# pair = ('LGM', 'glbs_full', 'glb', '+z')
 
 
 if __name__=="__main__":
-    input = {
-    "ply_file_url": "https://replicate.delivery/pbxt/UvKKgNj9mT7pIVHzwerhcjkp5cMH4FS5emPVghk2qyzMRwUSA/gradio_output.ply"
-    }
-    print("Start...")
-    output = replicate.run(
-    "camenduru/lgm-ply-to-glb:eb217314ab0d025370df16b8c9127f9ac1a0e4b3ffbff6b323d598d3c814d258",
-    input=input
-    )
-    print("output: ", output)
-    #=> "https://replicate.delivery/pbxt/r4iOSfk7cv2wACJL539ACB4E...
+    # input = {
+    # "input_image": "https://replicate.delivery/pbxt/KN0hQI9pYB3NOpHLqktkkQIblwpXt0IG7qI90n5hEnmV9kvo/bird_rgba.png",
+    # }
+    # print("Start...")
+    # model_client = replicate.Client(api_token=REPLICATE_API_TOKEN)
+    # output = model_client.run(
+    # "camenduru/lgm:d2870893aa115773465a823fe70fd446673604189843f39a99642dd9171e05e2",
+    # input=input
+    # )
+    # print("output: ", output)
+    #=>  ['https://replicate.delivery/pbxt/toffawxRE3h6AUofI9sPtiAsoYI0v73zuGDZjZWBWAPzHKSlA/gradio_output.mp4', 'https://replicate.delivery/pbxt/oSn1XPfoJuw2UKOUIAue2iXeT7aXncVjC4QwHKU5W5x0HKSlA/gradio_output.ply']
+
+    output = ['https://replicate.delivery/pbxt/RPSTEes37lzAJav3jy1lPuzizm76WGU4IqDcFcAMxhQocjUJA/gradio_output.mp4', 'https://replicate.delivery/pbxt/2Vy8yrPO3PYiI1YJBxPXAzryR0SC0oyqW3XKPnXiuWHUuRqE/gradio_output.ply']
+    to_mesh_client = Client("https://dylanebert-splat-to-mesh.hf.space/", upload_files=True, download_files=True)
+    mesh = to_mesh_client.predict(output[1], api_name="/run")
+    print(mesh)

requirements.txt

@@ -25,8 +25,6 @@ gradio_client==0.14.0
 h11==0.14.0
 httpcore==1.0.4
 httpx==0.27.0
-huggingface-cli==0.1
-huggingface-hub==0.22.0
 idna==3.6
 importlib_resources==6.4.0
 Jinja2==3.1.3

serve/gradio_web_i2s.py

@@ -43,7 +43,7 @@ Find out who is the 🥇conditional image generation models! More models are goi
 
 """
     model_list = models.get_i2s_models()
-    gen_func = partial(generate_i2s_multi_annoy, models.inference_parallel, models.render_parallel)
+    gen_func = partial(generate_i2s_multi_annoy, models.inference_parallel_anony, models.render_parallel)
 
     state_0 = gr.State()
     state_1 = gr.State()
@@ -140,28 +140,29 @@ Find out who is the 🥇conditional image generation models! More models are goi
     model_selectors = [model_selector_left, model_selector_right]
     results = [normal_left, rgb_left, normal_right, rgb_right]
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         leftvote_btn, rightvote_btn, tie_btn, bothbad_btn = btn_list
 
         leftvote_btn.click(
             leftvote_last_response_anony,
-            states + model_selectors,
-            [imagebox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         rightvote_btn.click(
             rightvote_last_response_anony,
-            states + model_selectors,
-            [imagebox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         tie_btn.click(
             tievote_last_response_anony,
-            states + model_selectors,
-            [imagebox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         bothbad_btn.click(
             bothbad_vote_last_response_anony,
-            states + model_selectors,
-            [imagebox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
 
     sample_btn.click(
@@ -172,6 +173,10 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     imagebox.upload(
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [imagebox] + model_selectors,
         states + results + model_selectors,
@@ -187,9 +192,9 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     send_btn.click(
-        sample_model,
-        states + [model_str],
-        states + model_selectors
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
     ).then(
         gen_func,
         states + [imagebox] + model_selectors,
@@ -221,9 +226,9 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     regenerate_btn.click(
-        sample_model,
-        states + [model_str],
-        states + model_selectors
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
     ).then(
         gen_func,
         states + [imagebox] + model_selectors,
@@ -379,28 +384,29 @@ Find out who is the 🥇conditional image generation models! More models are goi
         api_name="model_selector_right"
     )
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         leftvote_btn, rightvote_btn, tie_btn, bothbad_btn = btn_list
 
         leftvote_btn.click(
             leftvote_last_response_named,
-            states + model_selectors,
-            [imagebox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         rightvote_btn.click(
             rightvote_last_response_named,
-            states + model_selectors,
-            [imagebox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         tie_btn.click(
             tievote_last_response_named,
-            states + model_selectors,
-            [imagebox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         bothbad_btn.click(
             bothbad_vote_last_response_named,
-            states + model_selectors,
-            [imagebox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
 
     sample_btn.click(
@@ -411,9 +417,13 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     imagebox.upload(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [imagebox] + model_selectors,
-        states + results + model_selectors,
+        states + results,
         api_name="submit_btn_named"
     ).then(
         enable_mds,
@@ -426,9 +436,13 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     send_btn.click(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [imagebox] + model_selectors,
-        states + results + model_selectors,
+        states + results,
         api_name="send_btn_named"
     ).then(
         enable_mds,
@@ -456,6 +470,10 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     regenerate_btn.click(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [imagebox] + model_selectors,
         states + results + model_selectors,
@@ -487,7 +505,7 @@ def build_i2s_ui_single_model(models):
 
 """
     model_list = models.get_i2s_models()
-    gen_func = partial(generate_i2s, models.inference_parallel, models.render_parallel)
+    gen_func = partial(generate_i2s, models.inference, models.render)
 
     gr.Markdown(notice_markdown, elem_id="notice_markdown")
 
@@ -508,40 +526,39 @@ def build_i2s_ui_single_model(models):
         normal = gr.Image(width=512, label = "Normal", show_download_button=True)
         rgb = gr.Image(width=512, label = "RGB", show_download_button=True,)
 
-    with gr.Row():
-        imagebox = gr.Image(
-            width=512,
-            show_label=False,
-            visible=True,
-            interactive=True,
-            elem_id="input_box",
-        )
-        with gr.Column():
-            # with gr.Row():
-            sample_btn = gr.Button(value="🎲 Sample", variant="primary")
-            send_btn = gr.Button(value="📤 Send", variant="primary")
-            clear_btn = gr.Button(value="🗑️ Clear", interactive=False)
-            regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-
-
     with gr.Row(elem_id="Geometry Quality"):
-        gr.Markdown("Geometry Quality: ")
+        geo_md = gr.Markdown("Geometry Quality: ")
         geo_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         geo_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         geo_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
 
     with gr.Row(elem_id="Texture Quality"):
-        gr.Markdown("Texture Quality: ")
+        text_md = gr.Markdown("Texture Quality: ")
         text_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         text_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         text_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
     
     with gr.Row(elem_id="Alignment Quality"):
-        gr.Markdown("Alignment Quality: ")
+        align_md = gr.Markdown("Alignment Quality: ")
         align_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         align_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         align_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
 
+    with gr.Row():
+        imagebox = gr.Image(
+            width=512,
+            show_label=False,
+            visible=True,
+            interactive=True,
+            elem_id="input_box",
+        )
+        with gr.Column():
+            # with gr.Row():
+            sample_btn = gr.Button(value="🎲 Sample", variant="primary")
+            send_btn = gr.Button(value="📤 Send", variant="primary")
+            clear_btn = gr.Button(value="🗑️ Clear", interactive=False)
+            regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
+
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
     state = gr.State()
@@ -549,24 +566,25 @@ def build_i2s_ui_single_model(models):
     text_btn_list = [text_upvote_btn, text_downvote_btn, text_flag_btn]
     align_btn_list = [align_upvote_btn, align_downvote_btn, align_flag_btn]
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         upvote_btn, downvote_btn, flag_btn = btn_list
 
         upvote_btn.click(
             upvote_last_response,
-            [state, model_selector],
-            [imagebox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
             
         downvote_btn.click(
             downvote_last_response,
-            [state, model_selector],
-            [imagebox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
         flag_btn.click(
             flag_last_response,
-            [state, model_selector],
-            [imagebox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
 
     sample_btn.click(
@@ -577,6 +595,10 @@ def build_i2s_ui_single_model(models):
     )
 
     imagebox.upload(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, imagebox, model_selector],
         [state, normal, rgb],
@@ -589,6 +611,10 @@ def build_i2s_ui_single_model(models):
     )
 
     send_btn.click(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, imagebox, model_selector],
         [state, normal, rgb],
@@ -613,6 +639,10 @@ def build_i2s_ui_single_model(models):
     )
 
     regenerate_btn.click(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, imagebox, model_selector],
         [state, normal, rgb],

serve/gradio_web_t2i_anony.py

@@ -20,7 +20,7 @@ from .inference import(
     generate_t2s_multi,
     generate_t2s_multi_annoy
 )
-from .constants import TEXT_PROMPT_PATH
+from constants import TEXT_PROMPT_PATH
 
 with open(TEXT_PROMPT_PATH, 'r') as f:
     prompt_list = json.load(f)
@@ -41,7 +41,7 @@ Find out who is the 🥇conditional image generation models! More models are goi
 
 """
     model_list = models.get_t2s_models()
-    gen_func = partial(generate_t2s_multi_annoy, models.inference_parallel, models.render_parallel)
+    gen_func = partial(generate_t2s_multi_annoy, models.inference_parallel_anony, models.render_parallel)
 
     state_0 = gr.State()
     state_1 = gr.State()

serve/gradio_web_t2i_named.py

@@ -17,7 +17,7 @@ from .inference import(
     sample_prompt,
     generate_t2s_multi
 )
-from .constants import TEXT_PROMPT_PATH
+from constants import TEXT_PROMPT_PATH
 
 with open(TEXT_PROMPT_PATH, 'r') as f:
     prompt_list = json.load(f)

serve/gradio_web_t2i_single.py

@@ -11,7 +11,7 @@ from .inference import(
     sample_prompt,
     generate_t2s
 )
-from .constants import TEXT_PROMPT_PATH
+from constants import TEXT_PROMPT_PATH
 
 with open(TEXT_PROMPT_PATH, 'r') as f:
     prompt_list = json.load(f)

serve/gradio_web_t2s.py

@@ -43,7 +43,7 @@ Find out who is the 🥇conditional image generation models! More models are goi
 
 """
     model_list = models.get_t2s_models()
-    gen_func = partial(generate_t2s_multi_annoy, models.inference_parallel, models.render_parallel)
+    gen_func = partial(generate_t2s_multi_annoy, models.inference_parallel_anony, models.render_parallel)
 
 
     state_0 = gr.State()
@@ -135,28 +135,29 @@ Find out who is the 🥇conditional image generation models! More models are goi
     model_selectors = [model_selector_left, model_selector_right]
     results = [normal_left, rgb_left, normal_right, rgb_right]
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         leftvote_btn, rightvote_btn, tie_btn, bothbad_btn = btn_list
 
         leftvote_btn.click(
             leftvote_last_response_anony,
-            states + model_selectors,
-            [textbox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         rightvote_btn.click(
             rightvote_last_response_anony,
-            states + model_selectors,
-            [textbox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         tie_btn.click(
             tievote_last_response_anony,
-            states + model_selectors,
-            [textbox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
         bothbad_btn.click(
             bothbad_vote_last_response_anony,
-            states + model_selectors,
-            [textbox] + btn_list + model_selectors
+            states + model_selectors + [dim_md],
+            states + btn_list + model_selectors
         )
 
     sample_btn.click(
@@ -167,9 +168,9 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     textbox.submit(
-        sample_model,
-        states + [model_str],
-        states + model_selectors
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
     ).then(
         gen_func,
         states + [textbox] + model_selectors,
@@ -186,9 +187,9 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     send_btn.click(
-        sample_model,
-        states + [model_str],
-        states + model_selectors
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
     ).then(
         gen_func,
         states + [textbox] + model_selectors,
@@ -220,9 +221,9 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     regenerate_btn.click(
-        sample_model,
-        states + [model_str],
-        states + model_selectors
+        reset_states_side_by_side_anony,
+        states,
+        states + model_selectors + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
     ).then(
         gen_func,
         states + [textbox] + model_selectors,
@@ -379,28 +380,29 @@ Find out who is the 🥇conditional image generation models! More models are goi
         api_name="model_selector_right"
     )
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         leftvote_btn, rightvote_btn, tie_btn, bothbad_btn = btn_list
 
         leftvote_btn.click(
             leftvote_last_response_named,
-            states + model_selectors,
-            [textbox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         rightvote_btn.click(
             rightvote_last_response_named,
-            states + model_selectors,
-            [textbox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         tie_btn.click(
             tievote_last_response_named,
-            states + model_selectors,
-            [textbox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
         bothbad_btn.click(
             bothbad_vote_last_response_named,
-            states + model_selectors,
-            [textbox] + btn_list
+            states + model_selectors + [dim_md],
+            states + btn_list
         )
 
     sample_btn.click(
@@ -411,9 +413,13 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     textbox.submit(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [textbox] + model_selectors,
-        states + results + model_selectors,
+        states + results,
         api_name="submit_btn_named"
     ).then(
         enable_mds,
@@ -426,9 +432,13 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     send_btn.click(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [textbox] + model_selectors,
-        states + results + model_selectors,
+        states + results,
         api_name="send_btn_named"
     ).then(
         enable_mds,
@@ -456,9 +466,13 @@ Find out who is the 🥇conditional image generation models! More models are goi
     )
 
     regenerate_btn.click(
+        reset_states_side_by_side,
+        states,
+        states + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn, geo_md, text_md, align_md]
+    ).then(
         gen_func,
         states + [textbox] + model_selectors,
-        states + results + model_selectors,
+        states + results,
         api_name="regenerate_btn_named"
     ).then(
         enable_mds,
@@ -487,7 +501,7 @@ def build_t2s_ui_single_model(models):
 
 """
     model_list = models.get_t2s_models()
-    gen_func = partial(generate_t2s, models.inference_parallel, models.render_parallel)
+    gen_func = partial(generate_t2s, models.inference, models.render)
 
     gr.Markdown(notice_markdown, elem_id="notice_markdown")
 
@@ -507,41 +521,40 @@ def build_t2s_ui_single_model(models):
     with gr.Row():
         normal = gr.Image(width=512, label = "Normal", show_download_button=True)
         rgb = gr.Image(width=512, label = "RGB", show_download_button=True,)
-
-    with gr.Row():
-        textbox = gr.Textbox(
-            show_label=False,
-            placeholder="👉 Enter your prompt or Sample a random prompt, and press ENTER",
-            container=True,
-            elem_id="input_box",
-        )
-        sample_btn = gr.Button(value="🎲 Sample", variant="primary", scale=0)
-        send_btn = gr.Button(value="📤 Send", variant="primary", scale=0)
-
-    with gr.Row():
-        clear_btn = gr.Button(value="🗑️ Clear", interactive=False)
-        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-
     
     with gr.Row(elem_id="Geometry Quality"):
-        gr.Markdown("Geometry Quality: ", elem_id="evaldim_markdown")
+        geo_md = gr.Markdown("Geometry Quality: ", elem_id="evaldim_markdown")
         geo_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         geo_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         geo_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
 
     
     with gr.Row(elem_id="Texture Quality"):
-        gr.Markdown("Texture Quality: ", elem_id="evaldim_markdown")
+        text_md = gr.Markdown("Texture Quality: ", elem_id="evaldim_markdown")
         text_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         text_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         text_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
     
     with gr.Row(elem_id="Alignment Quality"):
-        gr.Markdown("Alignment Quality: ", elem_id="evaldim_markdown")
+        align_md =gr.Markdown("Alignment Quality: ", elem_id="evaldim_markdown")
         align_upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
         align_downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
         align_flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
 
+    with gr.Row():
+        textbox = gr.Textbox(
+            show_label=False,
+            placeholder="👉 Enter your prompt or Sample a random prompt, and press ENTER",
+            container=True,
+            elem_id="input_box",
+        )
+        sample_btn = gr.Button(value="🎲 Sample", variant="primary", scale=0)
+        send_btn = gr.Button(value="📤 Send", variant="primary", scale=0)
+
+    with gr.Row():
+        clear_btn = gr.Button(value="🗑️ Clear", interactive=False)
+        regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
+
     gr.Markdown(acknowledgment_md, elem_id="ack_markdown")
 
     state = gr.State()
@@ -549,24 +562,25 @@ def build_t2s_ui_single_model(models):
     text_btn_list = [text_upvote_btn, text_downvote_btn, text_flag_btn]
     align_btn_list = [align_upvote_btn, align_downvote_btn, align_flag_btn]
 
-    for btn_list in [geo_btn_list, text_btn_list, align_btn_list]:
+    for btn_list, dim_md in zip([geo_btn_list, text_btn_list, align_btn_list], 
+                                [geo_md, text_md, align_md]):
         upvote_btn, downvote_btn, flag_btn = btn_list
 
         upvote_btn.click(
             upvote_last_response,
-            [state, model_selector],
-            [textbox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
             
         downvote_btn.click(
             downvote_last_response,
-            [state, model_selector],
-            [textbox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
         flag_btn.click(
             flag_last_response,
-            [state, model_selector],
-            [textbox] + btn_list
+            [state, model_selector, dim_md],
+            [state] + btn_list
         )
 
     sample_btn.click(
@@ -577,6 +591,10 @@ def build_t2s_ui_single_model(models):
     )
 
     textbox.submit(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, textbox, model_selector],
         [state, normal, rgb],
@@ -589,6 +607,10 @@ def build_t2s_ui_single_model(models):
     )
 
     send_btn.click(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, textbox, model_selector],
         [state, normal, rgb],
@@ -613,6 +635,10 @@ def build_t2s_ui_single_model(models):
     )
 
     regenerate_btn.click(
+        reset_state,
+        state,
+        [state] + geo_btn_list + text_btn_list + align_btn_list + [regenerate_btn, clear_btn]
+    ).then(
         gen_func,
         [state, textbox, model_selector],
         [state, normal, rgb],

serve/inference.py

@@ -4,17 +4,27 @@ import time
 
 from .utils import *
 from .vote_utils import t2s_logger, t2s_multi_logger, i2s_logger, i2s_multi_logger
-from .constants import IMAGE_DIR, OFFLINE_DIR, TEXT_PROMPT_PATH
+from constants import IMAGE_DIR, OFFLINE_DIR, TEXT_PROMPT_PATH, IMAGE_PROMPT_PATH
 
 with open(TEXT_PROMPT_PATH, 'r') as f:
     prompt_list = json.load(f)
 
+with open(IMAGE_PROMPT_PATH, 'r') as f:
+    lines = f.readlines()
+
+image_list = {}
+for line in lines:
+    idx = line.split('.png')[0].split('_')[-1]
+    url = line.split(')')[0].split('(')[-1]
+    image_list[eval(idx)] = url
+
 
 class State:
     def __init__(self, 
                  model_name, i2s_mode=False, offline=False, 
                  prompt=None, image=None, offline_idx=None, 
-                 normal_video=None , rgb_video=None):
+                 normal_video=None , rgb_video=None,
+                 evaluted_dims=0):
         self.conv_id = uuid.uuid4().hex
         self.model_name = model_name
         self.i2s_mode = i2s_mode
@@ -27,15 +37,18 @@ class State:
         self.normal_video = normal_video
         self.rgb_video = rgb_video
 
+        self.evaluted_dims = evaluted_dims
+
     def dict(self):
         base = {
             "conv_id": self.conv_id,
             "model_name": self.model_name,
             "i2s_mode": self.i2s_mode,
             "offline": self.offline,
-            "prompt": self.prompt
+            "prompt": self.prompt,
+            "evaluted_dims": self.evaluted_dims,
             }
-        if not self.offline and not self.offline_idx:
+        if self.offline:
             base['offline_idx'] = self.offline_idx
         return base
 
@@ -91,6 +104,8 @@ def sample_prompt(state, model_name):
 
     state.model_name = model_name
     state.prompt = prompt
+    state.offline = True, 
+    state.offline_idx = idx
     return state, prompt
 
 def sample_prompt_side_by_side(state_0, state_1, model_name_0, model_name_1):
@@ -111,12 +126,14 @@ def sample_image(state, model_name):
     if state is None:
         state = State(model_name)
 
-    idx = random.randint(0, len(prompt_list)-1)
-    prompt = prompt_list[idx]
+    idx = random.sample(image_list.keys(), 1)[0]
+    img_url = image_list[idx]
 
     state.model_name = model_name
-    state.prompt = prompt
-    return state, prompt
+    state.image = img_url
+    state.offline = True, 
+    state.offline_idx = idx
+    return state, img_url
 
 def sample_image_side_by_side(state_0, state_1, model_name_0, model_name_1):
     if state_0 is None:
@@ -124,20 +141,21 @@ def sample_image_side_by_side(state_0, state_1, model_name_0, model_name_1):
     if state_1 is None:
         state_1 = State(model_name_1)
 
-    idx = random.randint(0, len(prompt_list)-1)
-    prompt = prompt_list[idx]
+
+    idx = random.sample(image_list.keys(), 1)[0]
+    img_url = image_list[idx]
 
     state_0.offline, state_1.offline = True, True
     state_0.offline_idx, state_1.offline_idx = idx, idx
-    state_0.prompt, state_1.prompt = prompt, prompt
-    return state_0, state_1, prompt
+    state_0.image, state_1.image = img_url, img_url
+    return state_0, state_1, img_url
 
 def generate_t2s(gen_func, render_func, 
                  state, 
                  text, 
                  model_name, 
                  request: gr.Request):
-    if not text:
+    if not text or text.strip()=="":
         raise gr.Warning("Prompt cannot be empty.")
     if not model_name:
         raise gr.Warning("Model name cannot be empty.")
@@ -145,11 +163,13 @@ def generate_t2s(gen_func, render_func,
     if state is None:
         state = State(model_name, i2s_mode=False, offline=False)
 
+    text = text.strip()
     ip = get_ip(request)
     t2s_logger.info(f"generate. ip: {ip}")
 
     state.model_name = model_name
     state.prompt = text
+    state.evaluted_dims = 0
     try:
         idx = prompt_list.index(text)
         state.offline = True
@@ -158,14 +178,15 @@ def generate_t2s(gen_func, render_func,
         state.offline = False
         state.offline_idx = None
 
-    if not state.offline and not state.offline_idx:
+    if state.offline and state.offline_idx:
         start_time = time.time()
-        normal_video = os.path.join(OFFLINE_DIR, "text2shape", model_name, "normal", f"{state.offline_idx}.mp4")
-        rgb_video = os.path.join(OFFLINE_DIR, "text2shape", model_name, "rgb", f"{state.offline_idx}.mp4")
+        videos = gen_func(text, model_name, offline=state.offline, offline_idx=state.offline_idx)
+        # normal_video = os.path.join(OFFLINE_DIR, "text2shape", model_name, "normal", f"{state.offline_idx}.mp4")
+        # rgb_video = os.path.join(OFFLINE_DIR, "text2shape", model_name, "rgb", f"{state.offline_idx}.mp4")
         
-        state.normal_video = normal_video
-        state.rgb_video = rgb_video
-        yield state, normal_video, rgb_video
+        state.normal_video = videos['normal']
+        state.rgb_video = videos['rgb']        
+        yield state, videos['normal'], videos['rgb']
 
         # logger.info(f"===output===: {output}")
         data = {
@@ -181,13 +202,13 @@ def generate_t2s(gen_func, render_func,
         shape = gen_func(text, model_name)
         generate_time = time.time() - start_time
 
-        normal_video, rgb_video = render_func(shape, model_name)
+        videos = render_func(shape, model_name)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
 
-        state.normal_video = normal_video
-        state.rgb_video = rgb_video        
-        yield state, normal_video, rgb_video
+        state.normal_video = videos['normal']
+        state.rgb_video = videos['rgb']    
+        yield state, videos['normal'], videos['rgb']
 
         # logger.info(f"===output===: {output}")
         data = {
@@ -217,7 +238,7 @@ def generate_t2s_multi(gen_func, render_func,
                        text, 
                        model_name_0, model_name_1, 
                        request: gr.Request):
-    if not text:
+    if not text or text.strip()=="":
         raise gr.Warning("Prompt cannot be empty.")
     if not model_name_0:
         raise gr.Warning("Model name A cannot be empty.")
@@ -229,11 +250,13 @@ def generate_t2s_multi(gen_func, render_func,
     if state_1 is None:
         state_1 = State(model_name_1, i2s_mode=False, offline=False)
     
+    text = text.strip()
     ip = get_ip(request)
     t2s_multi_logger.info(f"generate. ip: {ip}")
     
     state_0.model_name, state_1.model_name = model_name_0, model_name_1
     state_0.prompt, state_1.prompt = text, text
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
     try:
         idx = prompt_list.index(text)
         state_0.offline, state_1.offline = True, True
@@ -242,18 +265,17 @@ def generate_t2s_multi(gen_func, render_func,
         state_0.offline, state_1.offline = False, False
         state_0.offline_idx, state_1.offline_idx = None, None
 
-    if not state_0.offline and not state_0.offline_idx:
+    if state_0.offline and state_0.offline_idx:
         start_time = time.time()
-        normal_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
-        rgb_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
-        normal_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
-        rgb_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
+        videos_0, videos_1 = gen_func(text, model_name_0, model_name_1, offline=state_0.offline, offline_idx=state_0.offline_idx)
+        # normal_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
+        # rgb_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
+        # normal_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
+        # rgb_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
         
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1,videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb']
 
         # logger.info(f"===output===: {output}")
         data_0 = {
@@ -277,16 +299,13 @@ def generate_t2s_multi(gen_func, render_func,
         shape_0, shape_1 = gen_func(text, model_name_0, model_name_1)
         generate_time = time.time() - start_time
 
-        normal_video_0, rgb_video_0, normal_video_1, rgb_video_1 = render_func(shape_0, model_name_0,
-                                                                               shape_1, model_name_1)
+        videos_0, videos_1 = render_func(shape_0, model_name_0, shape_1, model_name_1)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
-
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1
+        
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb']
 
         # logger.info(f"===output===: {output}")
         data_0 = {
@@ -330,18 +349,20 @@ def generate_t2s_multi_annoy(gen_func, render_func,
                              text, 
                              model_name_0, model_name_1, 
                              request: gr.Request):
-    if not text:
+    if not text or text.strip()=="":
         raise gr.Warning("Prompt cannot be empty.")
     if state_0 is None:
         state_0 = State(model_name_0, i2s_mode=False, offline=False)
     if state_1 is None:
         state_1 = State(model_name_1, i2s_mode=False, offline=False)
     
+    text = text.strip()
     ip = get_ip(request)
     t2s_multi_logger.info(f"generate. ip: {ip}")
 
     state_0.model_name, state_1.model_name = model_name_0, model_name_1
     state_0.prompt, state_1.prompt = text, text
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
     try:
         idx = prompt_list.index(text)
         state_0.offline, state_1.offline = True, True
@@ -350,18 +371,19 @@ def generate_t2s_multi_annoy(gen_func, render_func,
         state_0.offline, state_1.offline = False, False
         state_0.offline_idx, state_1.offline_idx = None, None
 
-    if not state_0.offline and not state_0.offline_idx:
+    if state_0.offline and state_0.offline_idx:
         start_time = time.time()
-        normal_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
-        rgb_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
-        normal_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
-        rgb_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
+        videos_0, videos_1, model_name_0, model_name_1 = gen_func(text, model_name_0, model_name_1, 
+                                      i2s_model=False, offline=state_0.offline, offline_idx=state_0.offline_idx)
+        # normal_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
+        # rgb_video_0 = os.path.join(OFFLINE_DIR, "text2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
+        # normal_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
+        # rgb_video_1 = os.path.join(OFFLINE_DIR, "text2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
         
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_1, rgb_video_1, \
+        state_0.model_name, state_1.model_name = model_name_0, model_name_1
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb'], \
             gr.Markdown(f"### Model A: {model_name_0}"), gr.Markdown(f"### Model B: {model_name_1}")
 
         # logger.info(f"===output===: {output}")
@@ -383,19 +405,17 @@ def generate_t2s_multi_annoy(gen_func, render_func,
         }
     else:
         start_time = time.time()
-        shape_0, shape_1 = gen_func(text, model_name_0, model_name_1)
+        shape_0, shape_1, model_name_0, model_name_1 = gen_func(text, model_name_0, model_name_1)
         generate_time = time.time() - start_time
 
-        normal_video_0, rgb_video_0, normal_video_1, rgb_video_1 = render_func(shape_0, model_name_0,
-                                                                               shape_1, model_name_1)
+        videos_0, videos_1 = render_func(shape_0, model_name_0, shape_1, model_name_1)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
-
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1, \
+        
+        state_0.model_name, state_1.model_name = model_name_0, model_name_1
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0[0], videos_0[1], videos_1[0], videos_1[1], \
             gr.Markdown(f"### Model A: {model_name_0}"), gr.Markdown(f"### Model B: {model_name_1}")
 
         # logger.info(f"===output===: {output}")
@@ -437,7 +457,7 @@ def generate_t2s_multi_annoy(gen_func, render_func,
             
             
 def generate_i2s(gen_func, render_func, state, image, model_name, request: gr.Request):
-    if not image:
+    if image is None:
         raise gr.Warning("Image cannot be empty.")
     if not model_name:
         raise gr.Warning("Model name cannot be empty.")
@@ -449,15 +469,17 @@ def generate_i2s(gen_func, render_func, state, image, model_name, request: gr.Re
 
     state.model_name = model_name
     state.image = image
+    state.evaluted_dims = 0
 
-    if not state.offline and not state.offline_idx:
+    if state.offline and state.offline_idx:
         start_time = time.time()
-        normal_video = os.path.join(OFFLINE_DIR, "image2shape", model_name, "normal", f"{state.offline_idx}.mp4")
-        rgb_video = os.path.join(OFFLINE_DIR, "image2shape", model_name, "rgb", f"{state.offline_idx}.mp4")
+        videos = gen_func(image, model_name, offline=state.offline, offline_idx=state.offline_idx)
+        # normal_video = os.path.join(OFFLINE_DIR, "image2shape", model_name, "normal", f"{state.offline_idx}.mp4")
+        # rgb_video = os.path.join(OFFLINE_DIR, "image2shape", model_name, "rgb", f"{state.offline_idx}.mp4")
         
-        state.normal_video = normal_video
-        state.rgb_video = rgb_video
-        yield state, normal_video, rgb_video
+        state.normal_video = videos['normal']
+        state.rgb_video = videos['rgb']
+        yield state, videos['normal'], videos['rgb']
 
         # logger.info(f"===output===: {output}")
         data = {
@@ -473,13 +495,13 @@ def generate_i2s(gen_func, render_func, state, image, model_name, request: gr.Re
         shape = gen_func(image, model_name)
         generate_time = time.time() - start_time
 
-        normal_video, rgb_video = render_func(shape, model_name)
+        videos = render_func(shape, model_name)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
 
-        state.normal_video = normal_video
-        state.rgb_video = rgb_video        
-        yield state, normal_video, rgb_video
+        state.normal_video = videos['normal']
+        state.rgb_video = videos['rgb']
+        yield state, videos['normal'], videos['rgb']
 
         # logger.info(f"===output===: {output}")
         data = {
@@ -513,7 +535,7 @@ def generate_i2s_multi(gen_func, render_func,
                        image, 
                        model_name_0, model_name_1,
                        request: gr.Request):
-    if not image:
+    if image is None:
         raise gr.Warning("Image cannot be empty.")
     if not model_name_0:
         raise gr.Warning("Model name A cannot be empty.")
@@ -530,20 +552,19 @@ def generate_i2s_multi(gen_func, render_func,
 
     state_0.model_name, state_1.model_name = model_name_0, model_name_1
     state_0.image, state_1.image = image, image
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
 
-    if not state_0.offline and not state_0.offline_idx and \
-        not state_1.offline and not state_1.offline_idx:
+    if state_0.offline and state_0.offline_idx:
         start_time = time.time()
-        normal_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
-        rgb_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
-        normal_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
-        rgb_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
+        videos_0, videos_1 = gen_func(image, model_name_0, model_name_1, offline=state_0.offline, offline_idx=state_0.offline_idx)
+        # normal_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
+        # rgb_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
+        # normal_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
+        # rgb_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
         
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1, \
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb'], \
             gr.Markdown(f"### Model A: {model_name_0}"), gr.Markdown(f"### Model B: {model_name_1}")
 
         # logger.info(f"===output===: {output}")
@@ -568,16 +589,13 @@ def generate_i2s_multi(gen_func, render_func,
         shape_0, shape_1 = gen_func(image, model_name_0, model_name_1)
         generate_time = time.time() - start_time
 
-        normal_video_0, rgb_video_0, normal_video_1, rgb_video_1 = render_func(shape_0, model_name_0,
-                                                                               shape_1, model_name_1)
+        videos_0, videos_1 = render_func(shape_0, model_name_0, shape_1, model_name_1)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
 
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb']
 
         # logger.info(f"===output===: {output}")
         data_0 = {
@@ -621,12 +639,12 @@ def generate_i2s_multi(gen_func, render_func,
     #     save_image_file_on_log_server(output_file)
             
 
-def generate_i2s_multi_annoy(gen_func, 
+def generate_i2s_multi_annoy(gen_func, render_func,
                              state_0, state_1, 
                              image, 
                              model_name_0, model_name_1, 
                              request: gr.Request):
-    if not image:
+    if image is None:
         raise gr.Warning("Image cannot be empty.")
     if state_0 is None:
         state_0 = State(model_name_0, i2s_mode=True, offline=False)
@@ -638,20 +656,24 @@ def generate_i2s_multi_annoy(gen_func,
 
     state_0.model_name, state_1.model_name = model_name_0, model_name_1
     state_0.image, state_1.image = image, image
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
 
-    if not state_0.offline and not state_0.offline_idx and \
-        not state_1.offline and not state_1.offline_idx:
+    if state_0.offline and state_0.offline_idx and state_1.offline and state_1.offline_idx:
         start_time = time.time()
-        normal_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
-        rgb_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
-        normal_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
-        rgb_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
-        
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1, \
+        videos_0, videos_1, model_name_0, model_name_1 = gen_func(image, model_name_0, model_name_1, 
+                                      i2s_model=True, offline=state_0.offline, offline_idx=state_0.offline_idx)
+        # normal_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "normal", f"{state_0.offline_idx}.mp4")
+        # rgb_video_0 = os.path.join(OFFLINE_DIR, "image2shape", model_name_0, "rgb", f"{state_0.offline_idx}.mp4")
+        # normal_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "normal", f"{state_1.offline_idx}.mp4")
+        # rgb_video_1 = os.path.join(OFFLINE_DIR, "image2shape", model_name_1, "rgb", f"{state_1.offline_idx}.mp4")
+        print(state_0.dict())
+        print(state_1.dict())
+        print(videos_0)
+        print(videos_1)
+        state_0.model_name, state_1.model_name = model_name_0, model_name_1
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1, videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb'], \
             gr.Markdown(f"### Model A: {model_name_0}"), gr.Markdown(f"### Model B: {model_name_1}")
 
         # logger.info(f"===output===: {output}")
@@ -676,16 +698,13 @@ def generate_i2s_multi_annoy(gen_func,
         shape_0, shape_1 = gen_func(image, model_name_0, model_name_1)
         generate_time = time.time() - start_time
 
-        normal_video_0, rgb_video_0, normal_video_1, rgb_video_1 = render_func(shape_0, model_name_0,
-                                                                               shape_1, model_name_1)
+        videos_0, videos_1 = render_func(shape_0, model_name_0, shape_1, model_name_1)
         finish_time = time.time()
         render_time = finish_time - start_time - generate_time
 
-        state_0.normal_video = normal_video_0
-        state_0.rgb_video = rgb_video_0
-        state_1.normal_video = normal_video_1
-        state_1.rgb_video = rgb_video_1
-        yield state_0, state_1, normal_video_0, rgb_video_0, normal_video_0, rgb_video_1, \
+        state_0.normal_video, state_0.rgb_video = videos_0['normal'], videos_0['rgb']
+        state_1.normal_video, state_1.rgb_video = videos_1['normal'], videos_1['rgb']
+        yield state_0, state_1,  videos_0['normal'], videos_0['rgb'], videos_1['normal'], videos_1['rgb'], \
             gr.Markdown(f"### Model A: {model_name_0}"), gr.Markdown(f"### Model B: {model_name_1}")
 
         # logger.info(f"===output===: {output}")

serve/log_utils.py

@@ -14,7 +14,7 @@ from pathlib import Path
 
 import requests
 
-from .constants import LOGDIR, LOG_SERVER_ADDR, SAVE_LOG
+from constants import LOGDIR, LOG_SERVER_ADDR, SAVE_LOG
 from .utils import save_log_str_on_log_server
 
 

serve/utils.py

@@ -7,7 +7,7 @@ import numpy as np
 import gradio as gr
 from pathlib import Path
 from model.model_registry import *
-from .constants import LOGDIR, LOG_SERVER_ADDR, APPEND_JSON, SAVE_IMAGE, SAVE_LOG
+from constants import LOGDIR, LOG_SERVER_ADDR, APPEND_JSON, SAVE_IMAGE, SAVE_LOG
 from typing import Union
 
 
@@ -118,6 +118,29 @@ def enable_buttons():
 def disable_buttons():
     return tuple(gr.update(interactive=False) for _ in range(11))
 
+def reset_state(state):
+    state.normal_video, state.rgb_video = None, None
+    state.evaluted_dims = 0
+    return (state,) + tuple(gr.update(interactive=False) for _ in range(11))
+
+def reset_states_side_by_side(state_0, state_1):
+    state_0.normal_video, state_0.rgb_video = None, None
+    state_1.normal_video, state_1.rgb_video = None, None
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
+    return (state_0, state_1) \
+        + tuple(gr.update(visible=i>=12, interactive=False) for i in range(14)) \
+        + tuple(gr.update(visible=False) for _ in range(3))
+
+def reset_states_side_by_side_anony(state_0, state_1):
+    state_0.model_name, state_1.model_name = "", ""
+    state_0.normal_video, state_0.rgb_video = None, None
+    state_1.normal_video, state_1.rgb_video = None, None
+    state_0.evaluted_dims, state_1.evaluted_dims = 0, 0
+    return (state_0, state_1) \
+        + (gr.Markdown("", visible=False), gr.Markdown("", visible=False))\
+        + tuple(gr.update(visible=i>=12, interactive=False) for i in range(14)) \
+        + tuple(gr.update(visible=False) for _ in range(3))
+
 def clear_t2s_history():
     return None, "", None, None
    

serve/vote_utils.py

@@ -6,17 +6,18 @@ import gradio as gr
 from pathlib import Path
 from .utils import *
 from .log_utils import build_logger
-from .constants import IMAGE_DIR
+from constants import IMAGE_DIR
 
 t2s_logger = build_logger("gradio_web_server_text2shape", "gr_web_text2shape.log") # t2s = image generation, loggers for single model direct chat
 t2s_multi_logger = build_logger("gradio_web_server_text2shape_multi", "gr_web_text2shape_multi.log") # t2s_multi = image generation multi, loggers for side-by-side and battle
 i2s_logger = build_logger("gradio_web_server_image2shape", "gr_web_image2shape.log") # i2s = image editing, loggers for single model direct chat
 i2s_multi_logger = build_logger("gradio_web_server_image2shape_multi", "gr_web_image2shape_multi.log") # i2s_multi = image editing multi, loggers for side-by-side and battle
 
-def vote_last_response_t2s(state, vote_type, model_selector, request: gr.Request):
+def vote_last_response_t2s(state, dim, vote_type, model_selector, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
         data = {
             "tstamp": round(time.time(), 4),
+            "dim": dim,
             "type": vote_type,
             "model": model_selector,
             "state": state.dict(),
@@ -25,10 +26,11 @@ def vote_last_response_t2s(state, vote_type, model_selector, request: gr.Request
         fout.write(json.dumps(data) + "\n")
         append_json_item_on_log_server(data, get_conv_log_filename())
         
-def vote_last_response_t2s_multi(states, vote_type, model_selectors, request: gr.Request):
+def vote_last_response_t2s_multi(states, dim, vote_type, model_selectors, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
         data = {
             "tstamp": round(time.time(), 4),
+            "dim": dim,
             "type": vote_type,
             "models": [x for x in model_selectors],
             "states": [x.dict() for x in states],
@@ -42,10 +44,11 @@ def vote_last_response_t2s_multi(states, vote_type, model_selectors, request: gr
     #         state.output.save(f, 'PNG')
     #     save_image_file_on_log_server(output_file)
 
-def vote_last_response_i2s(state, vote_type, model_selector, request: gr.Request):
+def vote_last_response_i2s(state, dim, vote_type, model_selector, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
         data = {
             "tstamp": round(time.time(), 4),
+            "dim": dim,
             "type": vote_type,
             "model": model_selector,
             "state": state.dict(),
@@ -62,10 +65,11 @@ def vote_last_response_i2s(state, vote_type, model_selector, request: gr.Request
     # save_image_file_on_log_server(output_file)
     # save_image_file_on_log_server(source_file)
         
-def vote_last_response_i2s_multi(states, vote_type, model_selectors, request: gr.Request):
+def vote_last_response_i2s_multi(states, dim, vote_type, model_selectors, request: gr.Request):
     with open(get_conv_log_filename(), "a") as fout:
         data = {
             "tstamp": round(time.time(), 4),
+            "dim": dim,
             "type": vote_type,
             "models": [x for x in model_selectors],
             "states": [x.dict() for x in states],
@@ -85,209 +89,274 @@ def vote_last_response_i2s_multi(states, vote_type, model_selectors, request: gr
             
 
 ## Text-to-Shape Generation (t2s) Single Model Direct Chat
-def upvote_last_response_t2s(state, model_selector, request: gr.Request):
+def upvote_last_response_t2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    t2s_logger.info(f"upvote. ip: {ip}")
-    vote_last_response_t2s(state, "upvote", model_selector, request)
-    return ("",) + (disable_btn,) * 3
+    t2s_logger.info(f"upvote [{dim_md}]. ip: {ip}")
+    vote_last_response_t2s(state, dim_md, "upvote", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
-def downvote_last_response_t2s(state, model_selector, request: gr.Request):
+def downvote_last_response_t2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    t2s_logger.info(f"downvote. ip: {ip}")
-    vote_last_response_t2s(state, "downvote", model_selector, request)
-    return ("",) + (disable_btn,) * 3
+    t2s_logger.info(f"downvote [{dim_md}]. ip: {ip}")
+    vote_last_response_t2s(state, dim_md, "downvote", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
-def flag_last_response_t2s(state, model_selector, request: gr.Request):
+def flag_last_response_t2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    t2s_logger.info(f"flag. ip: {ip}")
-    vote_last_response_t2s(state, "flag", model_selector, request)
-    return ("",) + (disable_btn,) * 3
+    t2s_logger.info(f"flag [{dim_md}]. ip: {ip}")
+    vote_last_response_t2s(state, dim_md, "flag", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
 
 ## Text-to-Shape Generation Multi (t2s_multi) Side-by-Side and Battle
 def leftvote_last_response_t2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"leftvote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"leftvote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "leftvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "leftvote", [model_selector0, model_selector1], request
     )
-    return ("",) + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 def rightvote_last_response_t2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"rightvote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"rightvote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "rightvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "rightvote", [model_selector0, model_selector1], request
     )
-    return ("",) + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 def tievote_last_response_t2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"tievote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"tievote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "tievote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "tievote", [model_selector0, model_selector1], request
     )
-    return ("",) + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 def bothbad_vote_last_response_t2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"bothbad_vote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"bothbad_vote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "bothbad_vote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "bothbad_vote", [model_selector0, model_selector1], request
     )
-    return ("",) + (disable_btn,) * 4 
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 
 def leftvote_last_response_t2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"leftvote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"leftvote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "leftvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "leftvote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return ("",) + (disable_btn,) * 4 + names
+
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
 
 def rightvote_last_response_t2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"rightvote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"rightvote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "rightvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "rightvote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return ("",) + (disable_btn,) * 4 + names
+
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
 
 def tievote_last_response_t2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"tievote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"tievote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "tievote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "tievote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return ("",) + (disable_btn,) * 4 + names
 
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
+    
 def bothbad_vote_last_response_t2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    t2s_multi_logger.info(f"bothbad_vote (named). ip: {get_ip(request)}")
+    t2s_multi_logger.info(f"bothbad_vote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_t2s_multi(
-        [state0, state1], "bothbad_vote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "bothbad_vote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return ("",) + (disable_btn,) * 4 + names
+    
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
 
 ## Image-to-Shape (i2s) Single Model Direct Chat
-def upvote_last_response_i2s(state, model_selector, request: gr.Request):
+def upvote_last_response_i2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    i2s_logger.info(f"upvote. ip: {ip}")
-    vote_last_response_i2s(state, "upvote", model_selector, request)
-    return ("", "", gr.Image(height=512, width=512, type="pil"), "",) + (disable_btn,) * 3
+    i2s_logger.info(f"upvote [{dim_md}]. ip: {ip}")
+    vote_last_response_i2s(state, dim_md, "upvote", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
-def downvote_last_response_i2s(state, model_selector, request: gr.Request):
+def downvote_last_response_i2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    i2s_logger.info(f"downvote. ip: {ip}")
-    vote_last_response_i2s(state, "downvote", model_selector, request)
-    return ("", "", gr.Image(height=512, width=512, type="pil"), "",) + (disable_btn,) * 3
+    i2s_logger.info(f"downvote [{dim_md}]. ip: {ip}")
+    vote_last_response_i2s(state, dim_md, "downvote", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
-def flag_last_response_i2s(state, model_selector, request: gr.Request):
+def flag_last_response_i2s(state, model_selector, dim_md, request: gr.Request):
     ip = get_ip(request)
-    i2s_logger.info(f"flag. ip: {ip}")
-    vote_last_response_i2s(state, "flag", model_selector, request)
-    return ("", "", gr.Image(height=512, width=512, type="pil"), "",) + (disable_btn,) * 3
+    i2s_logger.info(f"flag [{dim_md}]. ip: {ip}")
+    vote_last_response_i2s(state, dim_md, "flag", model_selector, request)
+    state.evaluted_dims += 1
+    return (state,) + (disable_btn,) * 3
 
 
 ## Image-to-Shape Multi (i2s_multi) Side-by-Side and Battle
 def leftvote_last_response_i2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"leftvote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"leftvote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "leftvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "leftvote", [model_selector0, model_selector1], request
     )
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
         
 def rightvote_last_response_i2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"rightvote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"rightvote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "rightvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "rightvote", [model_selector0, model_selector1], request
     )
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 def tievote_last_response_i2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"tievote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"tievote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "tievote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "tievote", [model_selector0, model_selector1], request
     )
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 def bothbad_vote_last_response_i2s_named(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"bothbad_vote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"bothbad_vote [{dim_md}] (named). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "bothbad_vote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "bothbad_vote", [model_selector0, model_selector1], request
     )
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    return (state0, state1) + (disable_btn,) * 4
 
 
 def leftvote_last_response_i2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"leftvote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"leftvote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "leftvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "leftvote", [model_selector0, model_selector1], request
     )
-    # names = (
-    #     "### Model A: " + state0.model_name,
-    #     "### Model B: " + state1.model_name,
-    # )
-    # names = (state0.model_name, state1.model_name)
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4 + names
+
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
         
+
 def rightvote_last_response_i2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"rightvote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"rightvote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "rightvote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "rightvote", [model_selector0, model_selector1], request
     )
-    # names = (
-    #     "### Model A: " + state0.model_name,
-    #     "### Model B: " + state1.model_name,
-    # )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4 + names
+
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
+        
 
 def tievote_last_response_i2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"tievote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"tievote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "tievote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "tievote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4 + names
+    
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
+        
 
 def bothbad_vote_last_response_i2s_anony(
-    state0, state1, model_selector0, model_selector1, request: gr.Request
+    state0, state1, model_selector0, model_selector1, dim_md, request: gr.Request
 ):
-    i2s_multi_logger.info(f"bothbad_vote (anony). ip: {get_ip(request)}")
+    i2s_multi_logger.info(f"bothbad_vote [{dim_md}] (anony). ip: {get_ip(request)}")
     vote_last_response_i2s_multi(
-        [state0, state1], "bothbad_vote", [model_selector0, model_selector1], request
+        [state0, state1], dim_md, "bothbad_vote", [model_selector0, model_selector1], request
     )
-    names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
-    return gr.Image(height=512, width=512, type="pil") + (disable_btn,) * 4 + names
+    
+    state0.evaluted_dims += 1
+    state1.evaluted_dims += 1
+    if state0.evaluted_dims == state1.evaluted_dims == 3:
+        names = (gr.Markdown(f"### Model A: {state0.model_name}", visible=True), gr.Markdown(f"### Model B: {state1.model_name}", visible=True))
+        return (state0, state1) + (disable_btn,) * 4 + names
+    else:
+        return (state0, state1) + (disable_btn,) * 4 + ("", "")
+        
 
 
 share_js = """