Spaces:
Runtime error
Runtime error
import os | |
import sys | |
from datetime import datetime as Datetime | |
import gradio as gr | |
import json | |
import requests | |
import threading | |
import time | |
import hashlib | |
sys.path.append('./') | |
from oss_utils import * | |
from myconfigs import * | |
from cache_util import RedisCache | |
scale_depth = 0.7 | |
scale_pose = 0.5 | |
def md5_hash_file(filename): | |
"""Compute the MD5 hash of the contents of the given file.""" | |
# Create a new MD5 hash object | |
md5_hash = hashlib.md5() | |
# Open the file in binary mode and read chunks | |
with open(filename, 'rb') as f: | |
# Read and update hash string value in blocks of 4K | |
for byte_block in iter(lambda: f.read(4096), b""): | |
md5_hash.update(byte_block) | |
# Return the hexadecimal digest of the hash | |
return md5_hash.hexdigest() | |
def get_dirnames(filePath='', tail=".mp4", reserve_num=-1): | |
if not os.path.isdir(filePath): | |
return [] | |
if len(tail) <= 0: | |
return [] | |
len_tail = len(tail) | |
lists = os.listdir(filePath) | |
file_list = [] | |
for i in range(len(lists)): | |
cur_file = lists[i] | |
# if os.path.isfile(cur_file): | |
if len(cur_file) > 4 and cur_file[-len_tail:] == tail: | |
full_file_path = os.path.join(filePath, cur_file) | |
file_list.append(full_file_path) | |
sorted_list = sorted(file_list, key=lambda x: os.path.getctime(x)) | |
sorted_list = sorted_list[::-1] | |
if reserve_num > 0: | |
for i in range(len(sorted_list)): | |
if i >= reserve_num: | |
os.remove(sorted_list[i]) | |
return sorted_list | |
def sync_request_local(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
# "Authorization": APP_AUTH_TEXTURE, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
url_create_task = 'http://0.0.0.0:8000/api' | |
print(f"request_id: {request_id}, request type: video generation, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
res = json.loads(res_.content.decode()) | |
result_video_url = '' | |
if res['header']['status_name'] == 'Success': | |
result_video_url = res['payload']['output']['res_video_path'] | |
print(f"request_id: {request_id}, request type: video generation, retuen message: Succees, result: {result_video_url}") | |
else: | |
print(f"request_id: {request_id}, request type: video generation, retuen message: Faild, result: {result_video_url}") | |
return result_video_url | |
def sync_request_cartoon(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
"Authorization": EAS_AUTH_CARTOONRECOG, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
url_create_task = 'http://1096433202046721.cn-shanghai.pai-eas.aliyuncs.com/api/predict/videogene_supp/api' | |
print(f"request_id: {request_id}, request type: cartoon recognize, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
# print(res_) | |
# print(res_.content) | |
res = json.loads(res_.content.decode()) | |
cartoon_recog = '' | |
if res['payload']['output']['error_message'] == 'Success': | |
cartoon_recog = res['payload']['output']['key']['label'] | |
# print(f"request_id: {request_id} cartoon_recog: {cartoon_recog}") | |
print(f"request_id: {request_id}, request type: cartoon recognize, retuen message: Succees, result: {cartoon_recog}") | |
else: | |
print(f"request_id: {request_id}, request type: cartoon recognize, retuen message: Faild, result: {cartoon_recog}") | |
return cartoon_recog | |
def sync_request_translate_en2cn(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
"Authorization": EAS_AUTH_CARTOONRECOG, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
url_create_task = 'http://1096433202046721.cn-shanghai.pai-eas.aliyuncs.com/api/predict/videogene_supp/api' | |
print(f"[{request_id}], request type: translate en2cn, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
# print(res_) | |
# print(res_.content) | |
res = json.loads(res_.content.decode()) | |
translate_cn = '' | |
if res['payload']['output']['error_message'] == 'Success': | |
translate_cn = res['payload']['output']['key'] | |
# print(f"{request_id} translate_cn: {translate_cn}") | |
print(f"[{request_id}], request type: translate en2cn, retuen message: Succees, result: {translate_cn}") | |
else: | |
print(f"[{request_id}], request type: translate en2cn, retuen message: Faild, result: {translate_cn}") | |
return translate_cn | |
def async_request_and_query(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
# "Authorization": APP_AUTH_TEXTURE, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
# 1. async request | |
url_create_task = 'http://dashscope-scheduler-perf.1656375133437235.cn-beijing.pai-eas.aliyuncs.com/api/v1/task/submit' | |
print(f"request_id: {request_id}, request type: video generation, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
# print("res_=", res_) | |
# print(res_.content) | |
result_json = json.loads(res_.content.decode("utf-8")) | |
# 2.async query | |
is_running = True | |
running_print_count = 0 | |
res_video_path = None | |
while is_running: | |
url_query = 'http://dashscope-scheduler-perf.1656375133437235.cn-beijing.pai-eas.aliyuncs.com/api/v1/task/query-result' | |
res_ = requests.post(url_query, data=data, headers=headers) | |
respose_code = res_.status_code | |
if 200 == respose_code: | |
res = json.loads(res_.content.decode()) | |
if "SUCCESS" == res['header']['task_status']: | |
if 200 == res['payload']['output']['error_code']: | |
res_video_path = res['payload']['output']['res_video_path'] | |
print(f"request_id: {request_id}, request type: video generation, retuen message: Succees, result: {res_video_path}") | |
break | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {result_json}') | |
# raise gr.Error(f'algo error.') | |
break | |
elif "RUNNING" == res['header']['task_status']: | |
if running_print_count == 0: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: running..., result: {result_json}') | |
running_print_count += 1 | |
time.sleep(1) | |
elif "FAILED" == res['header']['task_status']: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {result_json}') | |
# raise gr.Error(f'query result faild.') | |
break | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {result_json}') | |
# raise gr.Error(f'query result faild.') | |
break | |
else: | |
print(f'request_id: {request_id}: Fail to query task result: {res_.content}') | |
# raise gr.Error("Fail to query task result.") | |
break | |
return res_video_path | |
def async_request_video_generation(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
# "Authorization": APP_AUTH_TEXTURE, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
# async request | |
url_create_task = 'http://dashscope-scheduler-perf.1656375133437235.cn-beijing.pai-eas.aliyuncs.com/api/v1/task/submit' | |
print(f"request_id: {request_id}, request type: video generation, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
# print("res_=", res_) | |
# print(res_.content) | |
# result_json = json.loads(res_.content.decode("utf-8")) | |
def async_query_video_generation(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
# "Authorization": APP_AUTH_TEXTURE, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
# async query | |
is_running = True | |
running_print_count = 0 | |
res_video_path = None | |
while is_running: | |
url_query = 'http://dashscope-scheduler-perf.1656375133437235.cn-beijing.pai-eas.aliyuncs.com/api/v1/task/query-result' | |
res_ = requests.post(url_query, data=data, headers=headers) | |
respose_code = res_.status_code | |
if 200 == respose_code: | |
res = json.loads(res_.content.decode()) | |
if "SUCCESS" == res['header']['task_status']: | |
if 200 == res['payload']['output']['error_code']: | |
res_video_path = res['payload']['output']['res_video_path'] | |
print(f"request_id: {request_id}, request type: video generation, retuen message: Succees, result: {res_video_path}") | |
break | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
# raise gr.Error(f'algo error.') | |
break | |
elif "RUNNING" == res['header']['task_status']: | |
if running_print_count == 0: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: running..., result: {res}') | |
running_print_count += 1 | |
time.sleep(1) | |
elif "FAILED" == res['header']['task_status']: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
break | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
break | |
else: | |
print(f'request_id: {request_id}: Fail to query task result: {res_.content}') | |
# raise gr.Error("Fail to query task result.") | |
break | |
return res_video_path | |
def query_video_generation(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
# "Authorization": APP_AUTH_TEXTURE, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
# async query | |
is_running = True | |
running_print_count = 0 | |
res_video_path = None | |
# while is_running: | |
url_query = 'http://dashscope-scheduler-perf.1656375133437235.cn-beijing.pai-eas.aliyuncs.com/api/v1/task/query-result' | |
res_ = requests.post(url_query, data=data, headers=headers) | |
respose_code = res_.status_code | |
res = json.loads(res_.content.decode()) | |
if 200 == respose_code: | |
if "SUCCESS" == res['header']['task_status']: | |
if 200 == res['payload']['output']['error_code']: | |
res_video_path = res['payload']['output']['res_video_path'] | |
print(f"request_id: {request_id}, request type: video generation, retuen message: Succees, result: {res_video_path}") | |
return "SUCCESS", res | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
return "FAILED", res | |
elif "RUNNING" == res['header']['task_status']: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: running..., result: {res}') | |
return "RUNNING", res | |
elif "FAILED" == res['header']['task_status']: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
return "FAILED", res | |
else: | |
print(f'request_id: {request_id}, request type: video generation, retuen message: Faild, result: {res}') | |
return "FAILED", res | |
else: | |
print(f'request_id: {request_id}: Fail to query task result: {res_.content}') | |
return "FAILED", res | |
def sync_request_prompt_caption(request_id, data): | |
headers = { | |
"Content-Type": "application/json", | |
"Accept": "application/json", | |
"Authorization": EAS_AUTH_PROMPT, | |
# "X-DashScope-Async": "enable", | |
# "X-DashScope-DataInspection": "enable" | |
} | |
url_create_task = 'http://1096433202046721.cn-shanghai.pai-eas.aliyuncs.com/api/predict/videogene_supp_gu50/api' | |
print(f"request_id: {request_id}, request type: prompt_caption, json input: {data}") | |
res_ = requests.post(url_create_task, data=data, headers=headers) | |
# print(res_) | |
# print(res_.content) | |
res = json.loads(res_.content.decode()) | |
prompt_caption = '' | |
gender = '' | |
if res['payload']['output']['error_message'] == 'Success': | |
key = res['payload']['output']['key'] # {"gender":"female","prompt":"Asian woman, 25-35, long black hair, dark brown eyes, average height","style":""} | |
gender = key['gender'] | |
prompt_caption = key['prompt'] | |
style = key['style'] | |
print(f"request_id: {request_id}, request type: prompt_caption, retuen message: Succees, result: {prompt_caption}") | |
else: | |
print(f"request_id: {request_id}, request type: prompt_caption, retuen message: Faild, result: {prompt_caption}") | |
return prompt_caption, gender, style | |
import re | |
def extract_mp4_filename(input_string): | |
# Regular expression pattern to match the file name | |
pattern = r"template_\d+\.mp4" | |
# Extract the MP4 file name | |
match = re.search(pattern, input_string) | |
if match: | |
return match.group() | |
else: | |
return None # or you could raise an exception or return an empty string | |
class HumanGenService: | |
def __init__(self): | |
self.oss_service = ossService() | |
self.all_user_requests = {} | |
self.all_requests = [] | |
self.all_requests_time = {} # dict: request_id, time | |
self.lock = threading.Lock() | |
def translate_en2cn(self, request_id, input_prompt): | |
#--------------- translate service -----------------# | |
translate_data = {} | |
translate_data['header'] = {} | |
translate_data['header']['request_id'] = request_id | |
translate_data['header']['service_id'] = '' | |
translate_data['header']['task_id'] = request_id | |
translate_data['header']['attributes'] = {} | |
translate_data['header']['attributes']['user_id'] = '' | |
translate_data['payload'] = {} | |
translate_data['payload']['input'] = {} | |
translate_data['payload']['input']['work_type'] = 'translate_en2zh' | |
translate_data['payload']['input']['key'] = input_prompt | |
translate_data['payload']['parameters'] = {} | |
translate_data = json.dumps(translate_data) # to string | |
# serving api | |
# print("input_prompt: ", input_prompt) | |
translate_cn = sync_request_translate_en2cn(request_id=request_id, data=translate_data) | |
# print("translate_cn: ", translate_cn) | |
#--------------- translate service -----------------# | |
print(f'[{request_id}] - [HumanGen] - translate ok') | |
return translate_cn | |
def click_button_prompt(self, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt='', prompt_template='',model_id=False): | |
cartoon_recog = 'realhuman' # input infer mode id 'face' and 'clothes' | |
if model_id is False: | |
cartoon_recog = 'realhuman' | |
elif model_id is True: | |
cartoon_recog = 'cartoon' | |
if input_mode == 'prompt_mode': | |
return input_prompt | |
if cartoon_recog == 'cartoon': | |
return prompt_template | |
print(f"request_id: {request_id}, input_mode: {input_mode}") | |
print(f"request_id: {request_id}, cartoon_recog: {cartoon_recog}") | |
print(f"request_id: {request_id}, ref_image_path: {ref_image_path}") | |
print(f"request_id: {request_id}, ref_video_path: {ref_video_path}") | |
if input_mode == 'template_mode' and cartoon_recog == 'realhuman' and (ref_video_path is None or ref_video_path == ''): | |
print(f"request_id: {request_id}, No video input, prompt caption over") | |
raise gr.Error("No video input!") | |
# return "请输入视频!" | |
ref_video_name = '' | |
user_mode = -1 | |
if input_mode == 'template_mode': | |
import re | |
# pattern = r"template_\d+\.mp4" | |
pattern = r"template_.*\.mp4" | |
# Extract the MP4 file name | |
match = re.search(pattern, ref_video_path) | |
if match: | |
ref_video_name = match.group() | |
user_mode = 0 # mode 0: image + template video | |
else: | |
user_mode = 1 # mode 1: image + upload video | |
if input_mode == 'prompt_mode': | |
user_mode = 2 # mode 2: image + prompt | |
if user_mode == 0 and cartoon_recog == 'realhuman': | |
return prompt_template | |
date_string = datetime.datetime.now().strftime('%Y-%m-%d') | |
img1_oss_path = '' | |
vid1_oss_path = '' | |
try: | |
if input_mode == 'template_mode' and cartoon_recog == 'realhuman': | |
vid_file_name = os.path.basename(ref_video_path) | |
vid_extension = os.path.splitext(vid_file_name)[1] # .mp4 | |
vid1_oss_path = self.oss_service.ObjectName + '/Service/' + date_string + '/' + user_id + '/' + request_id + '/' + "ref_video" + vid_extension | |
is_success1, sign_vid_oss_path = self.oss_service.uploadOssFile(vid1_oss_path, ref_video_path) | |
print(f"request_id: {request_id}, is_success1={is_success1}, sign_vid_oss_path={sign_vid_oss_path}") | |
except Exception as e: | |
print(f"request_id: {request_id}, oss upload error for input local image or video. ") | |
raise gr.Error("oss upload error for input local image or video") | |
#-----------------------------prompt caption-----------------------------# | |
# print("sign_img_oss_path: ", sign_img_oss_path) | |
# print("sign_img_oss_path1: ", sign_img_oss_path) | |
prompt_caption_data = {} | |
prompt_caption_data['header'] = {} | |
prompt_caption_data['header']['request_id'] = request_id | |
prompt_caption_data['header']['service_id'] = 'test123' | |
prompt_caption_data['header']['task_id'] = request_id | |
prompt_caption_data['header']['attributes'] = {} | |
prompt_caption_data['header']['attributes']['user_id'] = 'wanx_lab' | |
prompt_caption_data['payload'] = {} | |
prompt_caption_data['payload']['input'] = {} | |
prompt_caption_data['payload']['input']['work_type'] = 'prompt_caption' | |
sign_vid_oss_path = 'https://vigen-invi.oss-cn-shanghai-internal' + sign_vid_oss_path[len('https://vigen-invi.oss-cn-shanghai'):] | |
prompt_caption_data['payload']['input']['key'] = sign_vid_oss_path | |
prompt_caption_data['payload']['parameters'] = {} | |
prompt_caption_data['payload']['parameters']['input_type'] = 'reference_video' # id_image,cartoon_id_image,reference_video | |
prompt_caption_data['payload']['parameters']['input_format'] = 'url' # url,oss_path | |
prompt_caption_data = json.dumps(prompt_caption_data) # to string | |
# serving api | |
prompt_caption_en,__,__ = sync_request_prompt_caption(request_id=request_id, data=prompt_caption_data) | |
print(f"request_id: {request_id}, prompt caption: {prompt_caption_en}") | |
#-----------------------------prompt caption-----------------------------# | |
return prompt_caption_en # prompt_caption_cn | |
def template_video_2_prompt(self, ref_video_name): | |
file_name = ref_video_name | |
ref_ori_video_path = ref_video_name | |
print("videl filename:%s" % ref_video_name) | |
input_prompt = "" | |
if file_name[:len('template_')] == 'template_': | |
file_name = extract_mp4_filename(file_name) | |
if file_name not in template_prompts: | |
raise gr.Error("The input video is not a template!") | |
input_prompt = template_prompts[file_name] | |
if file_name not in template_orign_videos: | |
raise gr.Error("The input video is not a template!") | |
ref_ori_video_path = template_orign_videos[file_name] | |
else: | |
print("video file not found:%s" % ref_video_name) | |
return ref_ori_video_path, input_prompt | |
def click_button_func_async(self, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt='', prompt_template='',model_id=False): | |
start_time = time.time() | |
if is_wanx_platform: | |
user_id = 'wanx_lab' | |
request_id = get_random_string() | |
print(f"request_id: {request_id}, generate user_id: {user_id} and request_id: {request_id}") | |
if user_id is None or user_id == '': | |
user_id = 'test_version_phone' | |
if request_id is None or request_id == '': | |
request_id = get_random_string() | |
# key by: ref_video_name, digest(ref_image_path), prompt_template, input_prompt, | |
# scale_depth, scale_pose | |
#print("ref_image_path:%s ref_video_path:%s" % (ref_image_path, ref_video_path) ) | |
cache_key = None | |
# if ref_image_path and os.path.exists(ref_image_path): | |
# digest_ref_image = md5_hash_file(ref_image_path) | |
# if ref_video_path and os.path.exists(ref_video_path): | |
# ref_video_name = os.path.basename(ref_video_path) | |
# cache_key = "%s_%s_%s_%s_%s_%s" % (ref_video_name, digest_ref_image, prompt_template, input_prompt, scale_depth, scale_pose) | |
# print("cache key:%s" % cache_key) | |
# relative_oss_path = self.generate_video(cache_key, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt, prompt_template, model_id=model_id) | |
self.generate_video(cache_key, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt, prompt_template, model_id=model_id) | |
# oss_path = "oss://vigen-invi/" + relative_oss_path | |
# style = "video/snapshot,t_1000,f_jpg,w_544,h_768,m_fast" | |
# params = {'x-oss-process': style} | |
# _, snapshot_image = self.oss_service.sign(oss_path, timeout=3600*100, params=params) | |
# _, video_url = self.oss_service.sign(oss_path, timeout=3600*100) | |
total_time_minutes = (time.time() -start_time) / 60 # minites | |
print(f"request_id: {request_id}, Total process time: {total_time_minutes:.1f} minutes") | |
# return video_url, snapshot_image | |
#@RedisCache(expire=60*60*24*7) # 7 days | |
def generate_video(self, cache_key, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt='', prompt_template='',model_id=False): | |
self.lock.acquire() | |
if user_id in self.all_user_requests and len(self.all_user_requests[user_id]) > 0: | |
print(f"request_id: {request_id}, You still have unfinished tasks!") | |
self.lock.release() | |
raise gr.Error("You still have unfinished tasks!") | |
# return None | |
self.all_requests.append(request_id) | |
if request_id not in self.all_requests_time: | |
self.all_requests_time[request_id] = time.time() | |
if user_id not in self.all_user_requests: | |
self.all_user_requests[user_id] = [] | |
self.all_user_requests[user_id].append(request_id) | |
else: | |
self.all_user_requests[user_id].append(request_id) | |
self.lock.release() | |
print(f">>>request_id: {request_id}, user_id: {user_id}, add a new task.") | |
print(f"request_id: {request_id}, start process") | |
ref_video_name = '' | |
user_mode = -1 | |
if input_mode == 'template_mode': | |
import re | |
# pattern = r"template_\d+\.mp4" | |
pattern = r"template_.*\.mp4" | |
# Extract the MP4 file name | |
match = re.search(pattern, ref_video_path) | |
if match: | |
ref_video_name = match.group() | |
user_mode = 0 # mode 0: image + template video | |
else: | |
user_mode = 1 # mode 1: image + upload video | |
if input_mode == 'prompt_mode': | |
user_mode = 2 # mode 2: image + prompt | |
cartoon_recog = 'realhuman' # input infer mode id 'face' and 'clothes' | |
if model_id is False: | |
cartoon_recog = 'realhuman' | |
elif model_id is True: | |
cartoon_recog = 'cartoon' | |
# try: | |
# import shutil | |
# dir_path = "/tmp/gradio" | |
# shutil.rmtree(dir_path) | |
# print(f"目录: {dir_path} 删除成功") | |
# except OSError as e: | |
# print(f"目录: {dir_path} 删除失败:", e) | |
date_string = datetime.datetime.now().strftime('%Y-%m-%d') | |
img1_oss_path = '' | |
vid1_oss_path = '' | |
try: | |
# ref image | |
img_file_name = os.path.basename(ref_image_path) | |
img_extension = os.path.splitext(img_file_name)[1] # 输出:.jpg | |
img1_oss_path = self.oss_service.ObjectName + '/Service/' + date_string + '/' + user_id + '/' + request_id + '/' + "ref_image" + img_extension | |
is_success0, sign_img_oss_path = self.oss_service.uploadOssFile(img1_oss_path, ref_image_path) | |
print(f"request_id: {request_id}, is_success0={is_success0}, sign_img_oss_path={sign_img_oss_path}") | |
if user_mode == 0 or user_mode == 1: | |
vid_file_name = os.path.basename(ref_video_path) | |
vid_extension = os.path.splitext(vid_file_name)[1] # 输出:.mp4 | |
vid1_oss_path = self.oss_service.ObjectName + '/Service/' + date_string + '/' + user_id + '/' + request_id + '/' + "ref_video" + vid_extension | |
is_success1, sign_vid_oss_path = self.oss_service.uploadOssFile(vid1_oss_path, ref_video_path) | |
print(f"request_id: {request_id}, is_success1={is_success1}, sign_vid_oss_path={sign_vid_oss_path}") | |
except Exception as e: | |
print(f"request_id: {request_id}, Data upload faild!") | |
raise gr.Error("Data upload faild!") | |
# return None | |
#-----------------------------motion generation-----------------------------# | |
data = '{"header":{"request_id":"","service_id":"","task_id":""},"payload":{"input": {"ref_image_path": "", "ref_video_path": "", "ref_video_name": "", "input_prompt": "", "prompt_template": "", "scale_depth": 0.7, "scale_pose": 0.5},"parameters":{}}}' | |
data = json.loads(data) # string to dict | |
data['header']['service_id'] = DASHONE_SERVICE_ID | |
data['header']['request_id'] = request_id | |
data['header']['task_id'] = request_id | |
data['header']['attributes'] = {} | |
data['header']['attributes']['user_id'] = user_id | |
data['payload']['input']['user_id'] = user_id | |
data['payload']['input']['ref_image_path'] = img1_oss_path # sign_img_oss_path | |
data['payload']['input']['ref_video_path'] = vid1_oss_path # sign_vid_oss_path | |
data['payload']['input']['ref_video_name'] = ref_video_name | |
data['payload']['input']['input_prompt'] = input_prompt | |
data['payload']['input']['prompt_template'] = prompt_template | |
data['payload']['input']['scale_depth'] = scale_depth | |
data['payload']['input']['scale_pose'] = scale_pose | |
data['payload']['input']['cartoon_recog'] = cartoon_recog | |
data = json.dumps(data) # to string | |
# serving api | |
# sign_oss_path = sync_request_local(request_id=request_id, data=data) # sync | |
try: | |
async_request_video_generation(request_id=request_id, data=data) # async | |
print(f"request_id: {request_id}, async_request_video_generation") | |
except Exception as e: | |
print(f"request_id: {request_id}, The video generation task has failed, possibly due to a server connection issue, please try again.") | |
raise gr.Error("The video generation task has failed, possibly due to a server connection issue, please try again.") | |
# return None | |
print(f"=================end request_id: {request_id}") | |
# return sign_oss_path | |
def valid_check(self, user_id, request_id, input_mode, ref_image_path, ref_video_path, input_prompt='', prompt_template='',model_id=False): | |
if is_wanx_platform: | |
user_id = 'wanx_lab' | |
if user_id is None or user_id == '': | |
user_id = 'test_version_phone' | |
if request_id is None or request_id == '': | |
request_id = get_random_string() | |
print(f"-----------------request_id: {request_id}, user_id: {user_id}---------------") | |
self.lock.acquire() | |
if user_id in self.all_user_requests: | |
if len(self.all_user_requests[user_id]) > 0: | |
self.lock.release() | |
# raise gr.Error("Your video is in the process of being generated. Please wait until the process is complete before submitting a new task. Click refresh to get the latest progress updates.") | |
return "Your video is in the process of being generated. Please wait until the process is complete before submitting a new task. Click refresh to get the latest progress updates." | |
self.lock.release() | |
cartoon_recog = 'realhuman' # input infer mode id 'face' and 'clothes' | |
if model_id is False: | |
cartoon_recog = 'realhuman' | |
elif model_id is True: | |
cartoon_recog = 'cartoon' | |
print(f"request_id: {request_id}, input_mode: {input_mode}") | |
print(f"request_id: {request_id}, ref_image_path: {ref_image_path}") | |
print(f"request_id: {request_id}, ref_video_path: {ref_video_path}") | |
print(f"request_id: {request_id}, input_prompt = {input_prompt}") | |
print(f"request_id: {request_id}, prompt_template = {prompt_template}") | |
print(f"request_id: {request_id}, scale_depth = {scale_depth}") | |
print(f"request_id: {request_id}, scale_pose = {scale_pose}") | |
print(f"request_id: {request_id}, model_id = {model_id}") | |
print(f"request_id: {request_id}, style: {cartoon_recog}") | |
if ref_image_path is None or ref_image_path == '' or (not os.path.exists(ref_image_path)): | |
print(f"request_id: {request_id}, No image input, task over!") | |
# raise gr.Error("请输入图片!") | |
return "Please input a image." | |
if input_mode == 'template_mode' and cartoon_recog == 'realhuman' and (prompt_template == '' or prompt_template == [] or prompt_template is None): | |
print(f"request_id: {request_id}, No prompt input, task over!") | |
# raise gr.Error("请输入prompt!") | |
return "Please input a prompt." | |
if input_mode == 'prompt_mode' and cartoon_recog == 'realhuman' and (input_prompt == '' or input_prompt == [] or input_prompt is None): | |
print(f"request_id: {request_id}, No prompt input, task over!") | |
# raise gr.Error("请输入prompt!") | |
return "Please input a prompt." | |
if input_mode == 'template_mode' and (ref_video_path is None or ref_video_path == ''): | |
print(f"request_id: {request_id}, No video input, task over!") | |
# raise gr.Error("请输入视频!") | |
return "Please input a video." | |
ref_video_name = '' | |
user_mode = -1 | |
if input_mode == 'template_mode': | |
import re | |
# pattern = r"template_\d+\.mp4" | |
pattern = r"template_.*\.mp4" | |
# Extract the MP4 file name | |
match = re.search(pattern, ref_video_path) | |
if match: | |
ref_video_name = match.group() | |
user_mode = 0 # mode 0: image + template video | |
else: | |
user_mode = 1 # mode 1: image + upload video | |
if input_mode == 'prompt_mode': | |
user_mode = 2 # mode 2: image + prompt | |
print(f"request_id: {request_id}, user_mode = {user_mode} (0: image + template video, 1: image + upload video, 2: image + prompt)") | |
return '' | |
def delete_request_id(self, user_id, request_id, lock=True): | |
if lock: | |
self.lock.acquire() | |
if request_id in self.all_requests: | |
self.all_requests.remove(request_id) | |
if user_id in self.all_user_requests and request_id in self.all_user_requests[user_id]: | |
self.all_user_requests[user_id].remove(request_id) | |
if request_id in self.all_requests_time: | |
del self.all_requests_time[request_id] | |
if lock: | |
self.lock.release() | |
def get_ranking_location(self, user_id): | |
if is_wanx_platform: | |
user_id = 'wanx_lab' | |
if user_id is None or user_id == '': | |
user_id = 'test_version_phone' | |
process_status = '' | |
if len(self.all_requests) > 0: | |
for i in range(min(num_instance_dashone, len(self.all_requests))): | |
req = self.all_requests[i] | |
waste_time = 10000 | |
if req in self.all_requests_time: | |
endt = time.time() | |
startt = self.all_requests_time[req] | |
waste_time = (endt - startt)/60 | |
if waste_time > avg_process_time + 1: | |
data = '{"header":{"request_id":"","service_id":"","task_id":""},"payload":{"input": {"ref_image_path": "", "ref_video_path": "", "ref_video_name": "", "input_prompt": "", "prompt_template": "", "scale_depth": 0.7, "scale_pose": 0.5},"parameters":{}}}' | |
data = json.loads(data) # string to dict | |
data['header']['service_id'] = DASHONE_SERVICE_ID | |
data['header']['request_id'] = req | |
data['header']['task_id'] = req | |
data['header']['attributes'] = {} | |
data['header']['attributes']['user_id'] = user_id | |
data['payload']['input']['user_id'] = user_id | |
data = json.dumps(data) # to string | |
ret_status, ret_json = query_video_generation(request_id=req, data=data) | |
# print(f'ret_json = {ret_json}') | |
if ret_status == "SUCCESS" or ret_status == "FAILED": | |
if req in self.all_requests: | |
self.all_requests.remove(req) # delete request_id | |
if req in self.all_requests_time: | |
del self.all_requests_time[req] | |
else: | |
break | |
else: | |
break | |
else: | |
print(f'size of all_requests is empty.') | |
if user_id not in self.all_user_requests: | |
return f'You have not request a video generation task.', '' | |
if len(self.all_user_requests[user_id]) == 0: | |
return f'You do not have any tasks in progress, queue status.:{len(self.all_requests)}', '' | |
else: | |
self.lock.acquire() | |
lenn = len(self.all_user_requests[user_id]) | |
if lenn > 1: | |
for j in range(lenn - 1): | |
req = self.all_user_requests[user_id][j] | |
self.delete_request_id(user_id, req, lock=False) # delete request_id | |
request_id = self.all_user_requests[user_id][0] | |
self.lock.release() | |
data = '{"header":{"request_id":"","service_id":"","task_id":""},"payload":{"input": {"ref_image_path": "", "ref_video_path": "", "ref_video_name": "", "input_prompt": "", "prompt_template": "", "scale_depth": 0.7, "scale_pose": 0.5},"parameters":{}}}' | |
data = json.loads(data) # string to dict | |
data['header']['service_id'] = DASHONE_SERVICE_ID | |
data['header']['request_id'] = request_id | |
data['header']['task_id'] = request_id | |
data['header']['attributes'] = {} | |
data['header']['attributes']['user_id'] = user_id | |
data['payload']['input']['user_id'] = user_id | |
data = json.dumps(data) # to string | |
ret_status, ret_json = query_video_generation(request_id=request_id, data=data) | |
print(f'ret_json = {ret_json}') | |
if ret_status == "SUCCESS": | |
req = self.all_user_requests[user_id][0] | |
self.delete_request_id(user_id, req) # delete request_id | |
return 'Your video has been successfully generated.', '' | |
elif ret_status == "FAILED": | |
req = self.all_user_requests[user_id][0] | |
self.delete_request_id(user_id, req) # delete request_id | |
# if ret_json['header']['status_code'] == 200: | |
# if ret_json['header']['status_code'][] | |
return 'Your video generation failed, you may try enabling the "Cartoon Video Generation" option.', '' | |
else: | |
process_status = 'runing' | |
self.lock.acquire() | |
tmp_all_requests = self.all_requests.copy() | |
tmp_all_requests_time = self.all_requests_time.copy() | |
self.lock.release() | |
for i in range(len(tmp_all_requests)): | |
if tmp_all_requests[i] == request_id: | |
index = i + 1 | |
endt = time.time() | |
rest_time_list = [] | |
for k in range(i+1): | |
reqestid = tmp_all_requests[k] | |
startt = tmp_all_requests_time[reqestid] | |
wast_time = endt-startt | |
rest_time = max(0, avg_process_time * 60 - wast_time) | |
rest_time_list.append(rest_time) | |
sorted_time = sorted(rest_time_list, reverse=False) | |
print(f'rest_time_list: {rest_time_list}') | |
print(f'sorted_time: {sorted_time}') | |
process_time = 0 | |
if index <= num_instance_dashone: | |
process_time = rest_time_list[i]/60 | |
return f'Your video generation task is being processed, please wait for about {process_time:.1f} minutes, click refresh to get the latest progress update.', process_status | |
else: | |
rounds_to_wait = index // num_instance_dashone | |
rounds_to_rest = index % num_instance_dashone - 1 | |
process_time = rounds_to_wait * avg_process_time + sorted_time[rounds_to_rest]/60 | |
return f'There are {index-1} tasks ahead of you in the queue, please wait for about {process_time:.1f} minutes, click refresh to get the latest progress update.', process_status | |
return f'You do not have any tasks currently being processed.', process_status | |