Upload DATA/test_landmark.py with huggingface_hub
Browse files- DATA/test_landmark.py +158 -0
DATA/test_landmark.py
ADDED
@@ -0,0 +1,158 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# import os
|
2 |
+
# import cv2
|
3 |
+
# import time
|
4 |
+
# import glob
|
5 |
+
# import argparse
|
6 |
+
# import scipy
|
7 |
+
# import numpy as np
|
8 |
+
# from PIL import Image
|
9 |
+
# import torch
|
10 |
+
# from tqdm import tqdm
|
11 |
+
# from itertools import cycle
|
12 |
+
|
13 |
+
# from extract_kp_videos_safe import KeypointExtractor
|
14 |
+
# import numpy as np
|
15 |
+
# from PIL import Image
|
16 |
+
# with torch.no_grad():
|
17 |
+
# img_np =cv2.imread('Strawberry Monster.png')
|
18 |
+
# predictor = KeypointExtractor('cuda')
|
19 |
+
# dets = predictor.det_net.detect_faces(img_np, 0.97)
|
20 |
+
# if len(dets) == 0:
|
21 |
+
# detect = False
|
22 |
+
# else:
|
23 |
+
# print("success")
|
24 |
+
# import os
|
25 |
+
# import cv2
|
26 |
+
# import torch
|
27 |
+
# from tqdm import tqdm
|
28 |
+
# from extract_kp_videos_safe import KeypointExtractor
|
29 |
+
|
30 |
+
# # 创建 KeypointExtractor 实例
|
31 |
+
|
32 |
+
# # 设置文件夹路径
|
33 |
+
# folder_path = 'control_inversion'
|
34 |
+
# landmark_detect_false=0
|
35 |
+
# landmark_detect_success=0
|
36 |
+
# # 遍历文件夹中的图像文件
|
37 |
+
# for filename in tqdm(os.listdir(path)):
|
38 |
+
# if filename.endswith('.png') or filename.endswith('.jpg'):
|
39 |
+
# # 读取图像
|
40 |
+
# image_path = os.path.join(folder_path, filename)
|
41 |
+
# img_np = cv2.imread(image_path)
|
42 |
+
|
43 |
+
# # 进行人脸检测和关键点提取
|
44 |
+
# with torch.no_grad():
|
45 |
+
# predictor = KeypointExtractor('cuda')
|
46 |
+
# dets = predictor.det_net.detect_faces(img_np, 0.97)
|
47 |
+
# if len(dets) == 0:
|
48 |
+
# landmark_detect_false += 1
|
49 |
+
# else:
|
50 |
+
# landmark_detect_success += 1
|
51 |
+
|
52 |
+
# detect_rate = landmark_detect_success/(landmark_detect_success+landmark_detect_false)
|
53 |
+
# print(detect_rate)
|
54 |
+
# import os
|
55 |
+
# import cv2
|
56 |
+
# import torch
|
57 |
+
# from tqdm import tqdm
|
58 |
+
# from extract_kp_videos_safe import KeypointExtractor
|
59 |
+
|
60 |
+
# # 设置文件夹路径
|
61 |
+
# folder_path = 'prompts'
|
62 |
+
|
63 |
+
# # 初始化成功和失败的计数
|
64 |
+
# total_landmark_detect_success = 0
|
65 |
+
# total_landmark_detect_false = 0
|
66 |
+
|
67 |
+
# # 遍历文件夹中的 txt 文件
|
68 |
+
# for txt_filename in os.listdir(folder_path):
|
69 |
+
# if txt_filename.endswith('.txt'):
|
70 |
+
# txt_file_path = os.path.join(folder_path, txt_filename)
|
71 |
+
|
72 |
+
# # 读取 txt 文件中的图片列表
|
73 |
+
# with open(txt_file_path, 'r') as file:
|
74 |
+
# image_list = file.read().splitlines()
|
75 |
+
|
76 |
+
# landmark_detect_success = 0
|
77 |
+
# landmark_detect_false = 0
|
78 |
+
|
79 |
+
# # 遍历 txt 文件中的图片列表
|
80 |
+
# for image_filename in tqdm(image_list, desc=f'Processing {txt_filename}'):
|
81 |
+
# image_path = os.path.join('control_inversion', image_filename+'.png')
|
82 |
+
# if image_path.endswith('.png') or image_path.endswith('.jpg'):
|
83 |
+
# img_np = cv2.imread(image_path)
|
84 |
+
|
85 |
+
# # 进行人脸检测和关键点提取
|
86 |
+
# with torch.no_grad():
|
87 |
+
# predictor = KeypointExtractor('cuda')
|
88 |
+
# dets = predictor.det_net.detect_faces(img_np, 0.97)
|
89 |
+
|
90 |
+
# if len(dets) == 0:
|
91 |
+
# landmark_detect_false += 1
|
92 |
+
# else:
|
93 |
+
# landmark_detect_success += 1
|
94 |
+
|
95 |
+
# # 计算检测率
|
96 |
+
# detect_rate = landmark_detect_success / (landmark_detect_success + landmark_detect_false)
|
97 |
+
# print(f'{txt_filename}: Detect Rate = {detect_rate}')
|
98 |
+
|
99 |
+
# # 更新总的计数
|
100 |
+
# total_landmark_detect_success += landmark_detect_success
|
101 |
+
# total_landmark_detect_false += landmark_detect_false
|
102 |
+
|
103 |
+
# # 计算总的检测率
|
104 |
+
# total_detect_rate = total_landmark_detect_success / (total_landmark_detect_success + total_landmark_detect_false)
|
105 |
+
# print(f'Total Detect Rate = {total_detect_rate}')
|
106 |
+
import os
|
107 |
+
import sys
|
108 |
+
import cv2
|
109 |
+
import torch
|
110 |
+
from tqdm import tqdm
|
111 |
+
from chat_anything.sad_talker.face3d.extract_kp_videos_safe import KeypointExtractor
|
112 |
+
|
113 |
+
# 设置文件夹路径
|
114 |
+
folder_path = sys.argv[1]
|
115 |
+
|
116 |
+
# 初始化成功和失败的计数
|
117 |
+
total_landmark_detect_success = 0
|
118 |
+
total_landmark_detect_false = 0
|
119 |
+
|
120 |
+
# 遍历文件夹中的 txt 文件
|
121 |
+
for txt_filename in os.listdir(folder_path):
|
122 |
+
if txt_filename.endswith('.txt'):
|
123 |
+
txt_file_path = os.path.join(folder_path, txt_filename)
|
124 |
+
|
125 |
+
# # 读取 txt 文件中的图片列表
|
126 |
+
# with open(txt_file_path, 'r') as file:
|
127 |
+
# image_list = file.read().splitlines()
|
128 |
+
image_list = os.listdir(txt_file_path)
|
129 |
+
landmark_detect_success = 0
|
130 |
+
landmark_detect_false = 0
|
131 |
+
|
132 |
+
# 遍历 txt 文件中的图片列表
|
133 |
+
for image_filename in tqdm(image_list, desc=f'Processing {txt_filename}'):
|
134 |
+
image_path = os.path.join(txt_file_path, image_filename)
|
135 |
+
if image_path.endswith('.png') or image_path.endswith('.jpg'):
|
136 |
+
img_np = cv2.imread(image_path)
|
137 |
+
|
138 |
+
# 进行人脸检测和关键点提取
|
139 |
+
with torch.no_grad():
|
140 |
+
predictor = KeypointExtractor('cuda')
|
141 |
+
dets = predictor.det_net.detect_faces(img_np, 0.97)
|
142 |
+
|
143 |
+
if len(dets) == 0:
|
144 |
+
landmark_detect_false += 1
|
145 |
+
else:
|
146 |
+
landmark_detect_success += 1
|
147 |
+
|
148 |
+
# 计算检测率
|
149 |
+
detect_rate = landmark_detect_success / (landmark_detect_success + landmark_detect_false)
|
150 |
+
print(f'{txt_filename}: Detect Rate = {detect_rate}')
|
151 |
+
|
152 |
+
# 更新总的计数
|
153 |
+
total_landmark_detect_success += landmark_detect_success
|
154 |
+
total_landmark_detect_false += landmark_detect_false
|
155 |
+
|
156 |
+
# 计算总的检测率
|
157 |
+
total_detect_rate = total_landmark_detect_success / (total_landmark_detect_success + total_landmark_detect_false)
|
158 |
+
print(f'Total Detect Rate = {total_detect_rate}')
|