-
Notifications
You must be signed in to change notification settings - Fork 0
/
load_coco_face.py
239 lines (202 loc) · 8.25 KB
/
load_coco_face.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
"""
Process CMU Hand dataset to get cropped hand datasets.
"""
from tensorflow import keras
import json
import math
from tqdm import tqdm
import glob
import cv2
import numpy as np
from matplotlib import pyplot as plt
import os
import pandas as pd
from data_generator import generator2, generator3, generator3gray #generator2 single , generator3 multiple
from data_generator import generator22#, generator33 # binary mask generator22 single , generator33 multiple
NUM_KEYPOINTS = 68
from pycocotools.coco import COCO
# function iterates ofver all ocurrences of a person and returns relevant data row by row
def get_meta(coco):
ids = list(coco.imgs.keys())
for i, img_id in enumerate(ids):
img_meta = coco.imgs[img_id]
ann_ids = coco.getAnnIds(imgIds=img_id)
anns = coco.loadAnns(ann_ids)
img_file_name = img_meta['file_name']
w = img_meta['width']
h = img_meta['height']
url = img_meta['coco_url']
yield [img_id, img_file_name, w, h, url, anns]
def convert_to_df(coco, data_set):
images_data = []
persons_data = []
for img_id, img_fname, w, h, url, meta in get_meta(coco):
images_data.append({
'image_id': int(img_id),
'src_set_image_id': int(img_id), # repeat id to reference after join
'coco_url': url,
'path': data_set + '/' + img_fname,
'width': int(w),
'height': int(h)
})
for m in meta:
persons_data.append({
'ann_id': m['id'],
'image_id': m['image_id'],
'is_crowd': m['iscrowd'],
'face_valid': m['face_valid'],
'face_box': m['face_box'],
'face_kpts': m['face_kpts'],
'bbox': m['bbox'],
'bbox_area' : m['bbox'][2] * m['bbox'][3],
'face_box_area': m['face_box'][2] * m['face_box'][3],
'area': m['area'],
'num_keypoints': m['num_keypoints'],
'keypoints': m['keypoints'],
'segmentation': m['segmentation']
})
images_df = pd.DataFrame(images_data)
images_df.set_index('image_id', inplace=True)
persons_df = pd.DataFrame(persons_data)
persons_df.set_index('image_id', inplace=True)
return images_df, persons_df
def get_df(path_to_train_anns, path_to_val_anns):
train_coco = COCO(path_to_train_anns) # load annotations for training set
val_coco = COCO(path_to_val_anns) # load annotations for validation set
images_df, persons_df = convert_to_df(train_coco, 'train2017')
train_coco_df = pd.merge(images_df, persons_df, right_index=True, left_index=True)
train_coco_df['source'] = 0
train_coco_df.head()
images_df, persons_df = convert_to_df(val_coco, 'val2017')
val_coco_df = pd.merge(images_df, persons_df, right_index=True, left_index=True)
val_coco_df['source'] = 1
val_coco_df.head()
return pd.concat([train_coco_df, val_coco_df], ignore_index=True)
# ^ Dataframe containing all val and test keypoint annotations
def load_and_filter_annotations(path_to_train_anns, path_to_val_anns, subset):
df = get_df(path_to_train_anns, path_to_val_anns)
# apply filters here
print(f"Unfiltered df contains {len(df)} anns")
df = df.loc[df['is_crowd'] == 0] # drop crowd anns
df = df.loc[df['face_valid'] == 1] # drop crowd anns
KP_FILTERING_GT = 4 # Greater than x keypoints
df = df.loc[df['num_keypoints'] > KP_FILTERING_GT] # drop anns containing x kps
#BBOX_MIN_SIZE = 900 # Filter out images smaller than 30x30, TODO tweak
BBOX_MIN_SIZE = 100 # Filter out images smaller than 10x10, TODO tweak
#df = df.loc[df['bbox_area'] > BBOX_MIN_SIZE] # drop small bboxes
df = df.loc[df['face_box_area'] > BBOX_MIN_SIZE] # drop small bboxes
train_df = df.loc[df['source'] == 0]
val_df = df.loc[df['source'] == 1]
if subset != 1.0:
train_df = train_df.sample(frac=subset, random_state=1)
print(f"Train/Val dfs contains {len(train_df)}/{len(val_df)} anns")
return train_df.reset_index(), val_df.reset_index()
if __name__=="__main__":
train_annot_path = 'coco/annotations/coco_wholebody_train_v1.0.json'
val_annot_path = 'coco/annotations/coco_wholebody_val_v1.0.json'
train_img_path = 'coco/'
train_df, val_df = load_and_filter_annotations(train_annot_path, val_annot_path, subset=1.0)
x = generator22(train_df, train_img_path, shuffle=True, batch_size=16, input_dim=(256,256), output_dim=(256,256)) #cropped face
#x = generator3(train_df, train_img_path, shuffle=True, batch_size=16, input_dim=(256, 256), output_dim=(256, 256)) #resized uncropped face
#train_images, train_labels = generator(samples, batch_size=32, aug=None)#, aug=train_aug)
#print(len(x), len(y))
print("length of train genearator",len(x))
index = 0
for train_x,train_y in x:
print(train_x.shape, train_y.shape)
print(train_x[index].shape, train_y[index].shape)
break
cv2.imshow("data", train_x[index])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index])
plt.axis('off')
plt.savefig("cface1.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask1.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_x[index+1])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index+1])
plt.axis('off')
plt.savefig("cface2.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index+1].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index+1].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask2.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_x[index+2])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index+2])
plt.axis('off')
plt.savefig("cface3.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index+2].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index+2].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask3.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_x[index+3])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index+3])
plt.axis('off')
plt.savefig("cface4.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index+3].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index+3].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask4.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_x[index+4])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index+4])
plt.axis('off')
plt.savefig("cface5.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index+4].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index+4].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask5.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_x[index+5])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_x[index+5])
plt.axis('off')
plt.savefig("cface6.png", bbox_inches='tight', pad_inches=0)
plt.show()
cv2.imshow("data", train_y[index+5].sum(axis=-1))
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index+5].sum(axis=-1))
plt.axis('off')
plt.savefig("cmask6.png", bbox_inches='tight', pad_inches=0)
plt.show()
plt.imshow(train_y[index][:,:,0]) # 368,368
plt.show()
cv2.imshow("data", train_y[index][:,:,0])
cv2.waitKey(0)
cv2.destroyAllWindows()
plt.imshow(train_y[index][:,:,16]) # 368,368
plt.show()
cv2.imshow("data", train_y[index][:, :, 67])
cv2.waitKey(0)
cv2.destroyAllWindows()