support muliple checkpoint options.

app.py

@@ -1,9 +1,12 @@
 from cldm.ddim_hacked import DDIMSampler
 import math
 from omegaconf import OmegaConf
-from scripts.rendertext_tool import Render_Text, load_model_from_config
+from scripts.rendertext_tool import Render_Text, load_model_from_config, load_model_ckpt
 import gradio as gr  
 import os
+import torch
+import time
+
 def process_multi_wrapper(rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3,
                             shared_prompt,  
                             width_0, width_1, width_2, width_3,  
@@ -15,7 +18,7 @@ def process_multi_wrapper(rendered_txt_0, rendered_txt_1, rendered_txt_2, render
                             shared_num_samples, shared_image_resolution,  
                             shared_ddim_steps, shared_guess_mode,  
                             shared_strength, shared_scale, shared_seed,  
-                            shared_eta, shared_a_prompt, shared_n_prompt):  
+                            shared_eta, shared_a_prompt, shared_n_prompt, ):  
     
     rendered_txt_values = [rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3]  
     width_values = [width_0, width_1, width_2, width_3]  
@@ -66,23 +69,35 @@ def process_multi_wrapper_only_show_rendered(rendered_txt_0, rendered_txt_1, ren
                                      shared_eta, shared_a_prompt, shared_n_prompt, 
                                      only_show_rendered_image=True)  
 
+def load_ckpt(model_ckpt = "LAION-Glyph-10M"):
+    global render_tool, model
+    if torch.cuda.is_available():
+        for i in range(5):
+            torch.cuda.empty_cache()
+        time.sleep(2)
+        print("empty the cuda cache")
+
+    if model_ckpt == "LAION-Glyph-1M":
+        model = load_model_ckpt(model, "laion1M_model_wo_ema.ckpt")
+    elif model_ckpt == "LAION-Glyph-10M":
+        model = load_model_ckpt(model, "model_wo_ema.ckpt")
+    render_tool = Render_Text(model)
+    output_str = f"already change the model checkpoint to {model_ckpt}"
+    print(output_str)
+    if torch.cuda.is_available():
+        for i in range(5):
+            torch.cuda.empty_cache()
+        time.sleep(2)
+        print("empty the cuda cache")
 
 cfg = OmegaConf.load("config.yaml")
 model = load_model_from_config(cfg, "model_wo_ema.ckpt", verbose=True)
 # model = load_model_from_config(cfg, "model_states.pt", verbose=True)
 # model = load_model_from_config(cfg, "model.ckpt", verbose=True)
-
-ddim_sampler = DDIMSampler(model)
+# ddim_sampler = DDIMSampler(model)
 render_tool = Render_Text(model)
 
 
-# description = """
-# #  <center>Expedit-SAM (Expedite Segment Anything Model without any training)</center>
-# Github link: [Link](https://github.com/Expedit-LargeScale-Vision-Transformer/Expedit-SAM)
-# You can select the speed mode you want to use from the "Speed Mode" dropdown menu and click "Run" to segment the image you uploaded to the "Input Image" box.
-# Points per side is a hyper-parameter that controls the number of points used to generate the segmentation masks. The higher the number, the more accurate the segmentation masks will be, but the slower the inference speed will be. The default value is 12.
-# """
-
 description = """
 ## Control Stable Diffusion with Glyph Images
 """
@@ -121,8 +136,13 @@ with block:
                 with gr.Row():
                     run_button = gr.Button(value="Run")
                     show_render_button = gr.Button(value="Only Rendered")
+                with gr.Accordion("Model Options", open=False):
+                    with gr.Row():
+                        # model_ckpt = gr.inputs.Dropdown(["LAION-Glyph-10M", "Textcaps5K-10"], label="Checkpoint", default = "LAION-Glyph-10M")
+                        model_ckpt = gr.inputs.Dropdown(["LAION-Glyph-10M", "LAION-Glyph-1M"], label="Checkpoint", default = "LAION-Glyph-10M")
+                        load_button = gr.Button(value = "Load Checkpoint")
             
-            with gr.Accordion("Shared Advanced options", open=False):  
+            with gr.Accordion("Shared Advanced Options", open=False):  
                 with gr.Row():
                     shared_num_samples = gr.Slider(label="Images", minimum=1, maximum=12, value=1, step=1)  
                     shared_image_resolution = gr.Slider(label="Image Resolution", minimum=256, maximum=768, value=512, step=64, visible=False)  
@@ -170,7 +190,10 @@ with block:
                         shared_strength, shared_scale, shared_seed,  
                         shared_eta, shared_a_prompt, shared_n_prompt],  
                 outputs=[result_gallery]) 
-
-
+    
+    load_button.click(fn = load_ckpt,
+                inputs = [model_ckpt],
+                outputs = [result_gallery]
+    )
 
     block.launch()

app_old.py

@@ -0,0 +1,176 @@
+from cldm.ddim_hacked import DDIMSampler
+import math
+from omegaconf import OmegaConf
+from scripts.rendertext_tool import Render_Text, load_model_from_config
+import gradio as gr  
+import os
+def process_multi_wrapper(rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3,
+                            shared_prompt,  
+                            width_0, width_1, width_2, width_3,  
+                            ratio_0, ratio_1, ratio_2, ratio_3,  
+                            top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3,  
+                            top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3,  
+                            yaw_0, yaw_1, yaw_2, yaw_3,  
+                            num_rows_0, num_rows_1, num_rows_2, num_rows_3,  
+                            shared_num_samples, shared_image_resolution,  
+                            shared_ddim_steps, shared_guess_mode,  
+                            shared_strength, shared_scale, shared_seed,  
+                            shared_eta, shared_a_prompt, shared_n_prompt):  
+    
+    rendered_txt_values = [rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3]  
+    width_values = [width_0, width_1, width_2, width_3]  
+    ratio_values = [ratio_0, ratio_1, ratio_2, ratio_3]  
+    top_left_x_values = [top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3]  
+    top_left_y_values = [top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3]  
+    yaw_values = [yaw_0, yaw_1, yaw_2, yaw_3]  
+    num_rows_values = [num_rows_0, num_rows_1, num_rows_2, num_rows_3]  
+  
+    return render_tool.process_multi(rendered_txt_values, shared_prompt,  
+                                     width_values, ratio_values,  
+                                     top_left_x_values, top_left_y_values,  
+                                     yaw_values, num_rows_values,  
+                                     shared_num_samples, shared_image_resolution,  
+                                     shared_ddim_steps, shared_guess_mode,  
+                                     shared_strength, shared_scale, shared_seed,  
+                                     shared_eta, shared_a_prompt, shared_n_prompt 
+                                    ) 
+     
+def process_multi_wrapper_only_show_rendered(rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3,
+                            shared_prompt,  
+                            width_0, width_1, width_2, width_3,  
+                            ratio_0, ratio_1, ratio_2, ratio_3,  
+                            top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3,  
+                            top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3,  
+                            yaw_0, yaw_1, yaw_2, yaw_3,  
+                            num_rows_0, num_rows_1, num_rows_2, num_rows_3,  
+                            shared_num_samples, shared_image_resolution,  
+                            shared_ddim_steps, shared_guess_mode,  
+                            shared_strength, shared_scale, shared_seed,  
+                            shared_eta, shared_a_prompt, shared_n_prompt):  
+    
+    rendered_txt_values = [rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3]  
+    width_values = [width_0, width_1, width_2, width_3]  
+    ratio_values = [ratio_0, ratio_1, ratio_2, ratio_3]  
+    top_left_x_values = [top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3]  
+    top_left_y_values = [top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3]  
+    yaw_values = [yaw_0, yaw_1, yaw_2, yaw_3]  
+    num_rows_values = [num_rows_0, num_rows_1, num_rows_2, num_rows_3]  
+  
+    return render_tool.process_multi(rendered_txt_values, shared_prompt,  
+                                     width_values, ratio_values,  
+                                     top_left_x_values, top_left_y_values,  
+                                     yaw_values, num_rows_values,  
+                                     shared_num_samples, shared_image_resolution,  
+                                     shared_ddim_steps, shared_guess_mode,  
+                                     shared_strength, shared_scale, shared_seed,  
+                                     shared_eta, shared_a_prompt, shared_n_prompt, 
+                                     only_show_rendered_image=True)  
+
+
+cfg = OmegaConf.load("config.yaml")
+model = load_model_from_config(cfg, "model_wo_ema.ckpt", verbose=True)
+# model = load_model_from_config(cfg, "model_states.pt", verbose=True)
+# model = load_model_from_config(cfg, "model.ckpt", verbose=True)
+
+ddim_sampler = DDIMSampler(model)
+render_tool = Render_Text(model)
+
+
+# description = """
+# #  <center>Expedit-SAM (Expedite Segment Anything Model without any training)</center>
+# Github link: [Link](https://github.com/Expedit-LargeScale-Vision-Transformer/Expedit-SAM)
+# You can select the speed mode you want to use from the "Speed Mode" dropdown menu and click "Run" to segment the image you uploaded to the "Input Image" box.
+# Points per side is a hyper-parameter that controls the number of points used to generate the segmentation masks. The higher the number, the more accurate the segmentation masks will be, but the slower the inference speed will be. The default value is 12.
+# """
+
+description = """
+## Control Stable Diffusion with Glyph Images
+"""
+
+SPACE_ID = os.getenv('SPACE_ID')
+if SPACE_ID is not None:
+    # description += f'\n<p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. < a href=" ">< img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></ a></p >'
+    description += f'\n<p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. <a href="https://huggingface.co/spaces/{SPACE_ID}?duplicate=true"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></a></p>'
+
+block = gr.Blocks().queue()  
+
+with block:  
+    with gr.Row():  
+        gr.Markdown(description)  
+        only_show_rendered_image = gr.Number(value=1, visible=False)
+        
+    with gr.Column():  
+            
+        with gr.Row(): 
+            for i in range(4):  
+                with gr.Column():  
+                    exec(f"""rendered_txt_{i} = gr.Textbox(label=f"Render Text {i+1}")""")
+                    
+                    with gr.Accordion(f"Advanced options {i+1}", open=False):  
+                        exec(f"""width_{i} = gr.Slider(label="Bbox Width", minimum=0., maximum=1, value=0.3, step=0.01)  """)
+                        exec(f"""ratio_{i} = gr.Slider(label="Bbox_width_height_ratio", minimum=0., maximum=5, value=0., step=0.02, visible=False)  """)
+                        exec(f"""top_left_x_{i} = gr.Slider(label="Bbox Top Left x", minimum=0., maximum=1, value={0.35 - 0.25 * math.cos(math.pi * i)}, step=0.01)  """)
+                        exec(f"""top_left_y_{i} = gr.Slider(label="Bbox Top Left y", minimum=0., maximum=1, value={0.1 if i < 2 else 0.6}, step=0.01)  """)
+                        exec(f"""yaw_{i} = gr.Slider(label="Bbox Yaw", minimum=-180, maximum=180, value=0, step=5) """)
+                        # exec(f"""num_rows_{i} = gr.Slider(label="num_rows", minimum=1, maximum=4, value=1, step=1, visible=False)  """)
+                        exec(f"""num_rows_{i} = gr.Slider(label="num_rows", minimum=1, maximum=4, value=1, step=1)  """)
+        
+        with gr.Row(): 
+            with gr.Column():
+                shared_prompt = gr.Textbox(label="Shared Prompt")
+                with gr.Row():
+                    run_button = gr.Button(value="Run")
+                    show_render_button = gr.Button(value="Only Rendered")
+            
+            with gr.Accordion("Shared Advanced options", open=False):  
+                with gr.Row():
+                    shared_num_samples = gr.Slider(label="Images", minimum=1, maximum=12, value=1, step=1)  
+                    shared_image_resolution = gr.Slider(label="Image Resolution", minimum=256, maximum=768, value=512, step=64, visible=False)  
+                    shared_strength = gr.Slider(label="Control Strength", minimum=0.0, maximum=2.0, value=1.0, step=0.01, visible=False)  
+                    shared_guess_mode = gr.Checkbox(label='Guess Mode', value=False, visible=False)  
+                    shared_seed = gr.Slider(label="Seed", minimum=-1, maximum=2147483647, step=1, randomize=True)
+                with gr.Row():
+                    shared_scale = gr.Slider(label="Guidance Scale", minimum=0.1, maximum=30.0, value=9.0, step=0.1)  
+                    shared_ddim_steps = gr.Slider(label="Steps", minimum=1, maximum=100, value=20, step=1)    
+                    shared_eta = gr.Number(label="eta (DDIM)", value=0.0, visible=False)  
+                with gr.Row():
+                    shared_a_prompt = gr.Textbox(label="Added Prompt", value='best quality, extremely detailed')  
+                    shared_n_prompt = gr.Textbox(label="Negative Prompt",  
+                                            value='longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality') 
+        
+        with gr.Row(): 
+            result_gallery = gr.Gallery(label='Output', show_label=False, elem_id="gallery").style(grid=2, height='auto')  
+    
+    run_button.click(fn=process_multi_wrapper,  
+                inputs=[rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3,
+                        shared_prompt,  
+                        width_0, width_1, width_2, width_3,  
+                        ratio_0, ratio_1, ratio_2, ratio_3,  
+                        top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3,  
+                        top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3,  
+                        yaw_0, yaw_1, yaw_2, yaw_3,  
+                        num_rows_0, num_rows_1, num_rows_2, num_rows_3,  
+                        shared_num_samples, shared_image_resolution,  
+                        shared_ddim_steps, shared_guess_mode,  
+                        shared_strength, shared_scale, shared_seed,  
+                        shared_eta, shared_a_prompt, shared_n_prompt],  
+                outputs=[result_gallery])  
+    
+    show_render_button.click(fn=process_multi_wrapper_only_show_rendered,  
+                inputs=[rendered_txt_0, rendered_txt_1, rendered_txt_2, rendered_txt_3,
+                        shared_prompt,  
+                        width_0, width_1, width_2, width_3,  
+                        ratio_0, ratio_1, ratio_2, ratio_3,  
+                        top_left_x_0, top_left_x_1, top_left_x_2, top_left_x_3,  
+                        top_left_y_0, top_left_y_1, top_left_y_2, top_left_y_3,  
+                        yaw_0, yaw_1, yaw_2, yaw_3,  
+                        num_rows_0, num_rows_1, num_rows_2, num_rows_3,  
+                        shared_num_samples, shared_image_resolution,  
+                        shared_ddim_steps, shared_guess_mode,  
+                        shared_strength, shared_scale, shared_seed,  
+                        shared_eta, shared_a_prompt, shared_n_prompt],  
+                outputs=[result_gallery]) 
+
+
+
+    block.launch()

laion1M_model_wo_ema.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b86b22188bf580e80773a5ae101bf9787eb258349f3f1acf0ae50fd10cb3fec
+size 6671922039

scripts/rendertext_tool.py

@@ -46,6 +46,31 @@ def load_model_from_config(cfg, ckpt, verbose=False, not_use_ckpt=False):
     model.eval()
     return model
 
+def load_model_ckpt(model, ckpt, verbose=True):
+    map_location = "cpu" if not torch.cuda.is_available() else "cuda"
+    print("checkpoint map location:", map_location)
+    if ckpt.endswith("model_states.pt"):
+        sd = torch.load(ckpt, map_location=map_location)["module"]
+    else:
+        sd = load_state_dict(ckpt, location=map_location)
+   
+    keys_ = list(sd.keys())[:]
+    for k in keys_:
+        if k.startswith("module."):
+            nk = k[7:]
+            sd[nk] = sd[k]
+            del sd[k]
+
+    m, u = model.load_state_dict(sd, strict=False)
+    if len(m) > 0 and verbose:
+        print("missing keys: {}".format(len(m)))
+        print(m)
+    if len(u) > 0 and verbose:
+        print("unexpected keys: {}".format(len(u)))
+        print(u)
+    model.eval()
+    return model
+
 class Render_Text:
     def __init__(self, 
         model,