DATA/test_landmark.py

# import os
# import cv2
# import time
# import glob
# import argparse
# import scipy
# import numpy as np
# from PIL import Image
# import torch
# from tqdm import tqdm
# from itertools import cycle

# from extract_kp_videos_safe import KeypointExtractor
# import numpy as np
# from PIL import Image
# with torch.no_grad():
#     img_np =cv2.imread('Strawberry Monster.png')
#     predictor = KeypointExtractor('cuda')
#     dets = predictor.det_net.detect_faces(img_np, 0.97)
#     if len(dets) == 0:
#        detect = False
#     else:
#         print("success")
# import os
# import cv2
# import torch
# from tqdm import tqdm
# from extract_kp_videos_safe import KeypointExtractor

# # 创建 KeypointExtractor 实例

# # 设置文件夹路径
# folder_path = 'control_inversion'
# landmark_detect_false=0
# landmark_detect_success=0
# # 遍历文件夹中的图像文件
# for filename in tqdm(os.listdir(path)):
#     if filename.endswith('.png') or filename.endswith('.jpg'):
#         # 读取图像
#         image_path = os.path.join(folder_path, filename)
#         img_np = cv2.imread(image_path)

#         # 进行人脸检测和关键点提取
#         with torch.no_grad():
#             predictor = KeypointExtractor('cuda')
#             dets = predictor.det_net.detect_faces(img_np, 0.97)
#         if len(dets) == 0:
#             landmark_detect_false += 1
#         else:
#             landmark_detect_success += 1

# detect_rate = landmark_detect_success/(landmark_detect_success+landmark_detect_false)
# print(detect_rate)
# import os
# import cv2
# import torch
# from tqdm import tqdm
# from extract_kp_videos_safe import KeypointExtractor

# # 设置文件夹路径
# folder_path = 'prompts'

# # 初始化成功和失败的计数
# total_landmark_detect_success = 0
# total_landmark_detect_false = 0

# # 遍历文件夹中的 txt 文件
# for txt_filename in os.listdir(folder_path):
#     if txt_filename.endswith('.txt'):
#         txt_file_path = os.path.join(folder_path, txt_filename)

#         # 读取 txt 文件中的图片列表
#         with open(txt_file_path, 'r') as file:
#             image_list = file.read().splitlines()

#         landmark_detect_success = 0
#         landmark_detect_false = 0

#         # 遍历 txt 文件中的图片列表
#         for image_filename in tqdm(image_list, desc=f'Processing {txt_filename}'):
#             image_path = os.path.join('control_inversion', image_filename+'.png')
#             if image_path.endswith('.png') or image_path.endswith('.jpg'):
#                 img_np = cv2.imread(image_path)

#                 # 进行人脸检测和关键点提取
#                 with torch.no_grad():
#                     predictor = KeypointExtractor('cuda')
#                     dets = predictor.det_net.detect_faces(img_np, 0.97)

#                 if len(dets) == 0:
#                     landmark_detect_false += 1
#                 else:
#                     landmark_detect_success += 1

#         # 计算检测率
#         detect_rate = landmark_detect_success / (landmark_detect_success + landmark_detect_false)
#         print(f'{txt_filename}: Detect Rate = {detect_rate}')

#         # 更新总的计数
#         total_landmark_detect_success += landmark_detect_success
#         total_landmark_detect_false += landmark_detect_false

# # 计算总的检测率
# total_detect_rate = total_landmark_detect_success / (total_landmark_detect_success + total_landmark_detect_false)
# print(f'Total Detect Rate = {total_detect_rate}')
import os
import sys
import cv2
import torch
from tqdm import tqdm
from chat_anything.sad_talker.face3d.extract_kp_videos_safe import KeypointExtractor

# 设置文件夹路径
folder_path = sys.argv[1]

# 初始化成功和失败的计数
total_landmark_detect_success = 0
total_landmark_detect_false = 0

# 遍历文件夹中的 txt 文件
for txt_filename in os.listdir(folder_path):
    if txt_filename.endswith('.txt'):
        txt_file_path = os.path.join(folder_path, txt_filename)

        # # 读取 txt 文件中的图片列表
        # with open(txt_file_path, 'r') as file:
        #     image_list = file.read().splitlines()
        image_list = os.listdir(txt_file_path)
        landmark_detect_success = 0
        landmark_detect_false = 0

        # 遍历 txt 文件中的图片列表
        for image_filename in tqdm(image_list, desc=f'Processing {txt_filename}'):
            image_path = os.path.join(txt_file_path, image_filename)
            if image_path.endswith('.png') or image_path.endswith('.jpg'):
                img_np = cv2.imread(image_path)

                # 进行人脸检测和关键点提取
                with torch.no_grad():
                    predictor = KeypointExtractor('cuda')
                    dets = predictor.det_net.detect_faces(img_np, 0.97)

                if len(dets) == 0:
                    landmark_detect_false += 1
                else:
                    landmark_detect_success += 1

        # 计算检测率
        detect_rate = landmark_detect_success / (landmark_detect_success + landmark_detect_false)
        print(f'{txt_filename}: Detect Rate = {detect_rate}')

        # 更新总的计数
        total_landmark_detect_success += landmark_detect_success
        total_landmark_detect_false += landmark_detect_false

# 计算总的检测率
total_detect_rate = total_landmark_detect_success / (total_landmark_detect_success + total_landmark_detect_false)
print(f'Total Detect Rate = {total_detect_rate}')