-
Notifications
You must be signed in to change notification settings - Fork 47
/
data_processed.py
209 lines (192 loc) · 9.23 KB
/
data_processed.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
import torch
from torchvision import transforms
import copy
import random
import os
import numpy as np
from PIL import Image
from imageio import imread
from torch.utils.data import Dataset
from torchvision.transforms import Compose, CenterCrop, ToTensor, Resize
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torchvision.transforms as transforms
import cv2
def filt_small_instance(coco_item, pixthreshold=4000,imgNthreshold=5):
list_dict = coco_item.catToImgs
for catid in list_dict:
list_dict[catid] = list(set( list_dict[catid] ))
new_dict = copy.deepcopy(list_dict)
for catid in list_dict:
imgids = list_dict[catid]
for n in range(len(imgids)):
imgid = imgids[n]
anns = coco_item.imgToAnns[imgid]
has_large_instance = False
for ann in anns:
if (ann['category_id'] == catid) and (ann['iscrowd'] == 0) and (ann['area'] > pixthreshold):
has_large_instance = True
if has_large_instance is False:
new_dict[catid].remove(imgid)
imgN = len(new_dict[catid])
if imgN <imgNthreshold:
new_dict.pop(catid)
print('catid:%d remain %d images, delet it!'%(catid,imgN))
else:
print('catid:%d remain %d images' % (catid, imgN))
print('remain %d categories'%len(new_dict))
np.save('./utils/new_cat2imgid_dict%d.npy'%pixthreshold, new_dict)
return new_dict
def train_data_producer(coco_item, datapath, npy, q, batch_size=10, group_size=5, img_size=224):
img_transform = transforms.Compose([transforms.Resize((img_size, img_size)), transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
gt_transform = transforms.Compose([transforms.Resize((img_size, img_size)), transforms.ToTensor()])
img_transform_gray = transforms.Compose([transforms.Resize((img_size, img_size)), transforms.ToTensor(),
transforms.Normalize(mean=[0.449], std=[0.226])])
if os.path.exists(npy):
# list_dict = np.load(npy).item()
list_dict = np.load(npy, allow_pickle=True).item()
else:
list_dict = filt_small_instance(coco_item, pixthreshold=4000, imgNthreshold=100)
catid2label={}
n=0
for catid in list_dict:
catid2label[catid] = n
n=n+1
while 1:
rgb = torch.zeros(batch_size*group_size, 3, img_size, img_size)
cls_labels = torch.zeros(batch_size, 78)
mask_labels = torch.zeros(batch_size*group_size, img_size, img_size)
if batch_size> len(list_dict):
remainN = batch_size - len(list_dict)
batch_catid = random.sample(list_dict.keys(), remainN) + random.sample(list_dict, len(list_dict))
else:
batch_catid = random.sample(list_dict.keys(), batch_size)
group_n = 0
img_n = 0
for catid in batch_catid:
imgids = random.sample(list_dict[catid], group_size)
co_catids = []
anns = coco_item.imgToAnns[imgids[0]]
for ann in anns:
if (ann['iscrowd'] == 0) and (ann['area'] > 4000):
co_catids.append(ann['category_id'])
co_catids_backup = copy.deepcopy(co_catids)
for imgid in imgids[1:]:
img_catids = []
anns = coco_item.imgToAnns[imgid]
for ann in anns:
if (ann['iscrowd'] == 0) and (ann['area'] > 4000):
img_catids.append(ann['category_id'])
for co_catid in co_catids_backup:
if co_catid not in img_catids:
co_catids.remove(co_catid)
co_catids_backup = copy.deepcopy(co_catids)
for co_catid in co_catids:
cls_labels[group_n, catid2label[co_catid]] = 1
for imgid in imgids:
path = datapath + '%012d.jpg'%imgid
img = Image.open(path)
if img.mode == 'RGB':
img = img_transform(img)
else:
img = img_transform_gray(img)
anns = coco_item.imgToAnns[imgid]
mask = None
for ann in anns:
if ann['category_id'] in co_catids:
if mask is None:
mask = coco_item.annToMask(ann)
else:
mask = mask + coco_item.annToMask(ann)
mask[mask > 0] = 255
mask = Image.fromarray(mask)
mask = gt_transform(mask)
mask[mask > 0.5] = 1
mask[mask <= 0.5] = 0
rgb[img_n,:,:,:] = copy.deepcopy(img)
mask_labels[img_n,:,:] = copy.deepcopy(mask)
img_n = img_n + 1
group_n = group_n + 1
idx = mask_labels[:, :, :] > 1
mask_labels[idx] = 1
q.put([rgb, cls_labels, mask_labels])
def img_normalize(image):
if len(image.shape)==2:
channel = (image[:, :, np.newaxis] - 0.485) / 0.229
image = np.concatenate([channel,channel,channel], axis=2)
else:
image = (image-np.array([0.485, 0.456, 0.406], dtype=np.float32).reshape((1, 1, 3)))\
/np.array([0.229, 0.224, 0.225], dtype=np.float32).reshape((1, 1, 3))
return image
davis_fbms=['bear', 'bear01', 'bear02', 'bmx-bumps', 'boat', 'breakdance-flare', 'bus', 'car-turn', 'cars2', 'cars3', 'cars6', 'cars7', 'cars8', 'cars9', 'cats02', 'cats04', 'cats05', 'cats07', 'dance-jump', 'dog-agility', 'drift-turn', 'ducks01', 'elephant', 'flamingo', 'hike', 'hockey', 'horsejump-low', 'horses01', 'horses03', 'horses06', 'kite-walk', 'lion02', 'lucia', 'mallard-fly', 'mallard-water', 'marple1', 'marple10', 'marple11', 'marple13', 'marple3', 'marple5', 'marple8', 'meerkats01', 'motocross-bumps', 'motorbike', 'paragliding', 'people04', 'people05', 'rabbits01', 'rabbits05', 'rhino', 'rollerblade', 'scooter-gray', 'soccerball', 'stroller', 'surf', 'swing', 'tennis', 'train']
class VideoDataset(Dataset):
def __init__(self, dir_,epochs, size=224, group=5, use_flow=False):
self.img_list=[]
self.label_list=[]
self.flow_list=[]
self.group=group
dir_img=os.path.join(dir_,'image')
dir_gt=os.path.join(dir_,'groundtruth')
dir_flow=os.path.join(dir_,'flow')
self.dir_list=sorted(os.listdir(dir_img))
self.leng=0
for i in range(len(self.dir_list)):
ok=0
if self.dir_list[i] in davis_fbms:
ok=1
if ok==0:
continue
tmp_list=[]
cur_dir=sorted(os.listdir(os.path.join(dir_img,self.dir_list[i])))
for j in range(len(cur_dir)):
tmp_list.append(os.path.join(dir_img,self.dir_list[i],cur_dir[j]))
self.leng+=len(tmp_list)
self.img_list.append(tmp_list)
tmp_list=[]
cur_dir=sorted(os.listdir(os.path.join(dir_gt,self.dir_list[i])))
for j in range(len(cur_dir)):
tmp_list.append(os.path.join(dir_gt,self.dir_list[i],cur_dir[j]))
self.label_list.append(tmp_list)
self.img_size=224
self.dataset_len = epochs
self.use_flow=use_flow
self.dir_=dir_
def __len__(self):
return self.dataset_len
def __getitem__(self, item):
rd=np.random.randint(0,len(self.img_list))
rd2=np.random.permutation(len(self.img_list[rd]))
cur_img=[]
cur_flow=[]
cur_gt=[]
for i in range(self.group):
cur_img.append(self.img_list[rd][rd2[i%len(self.img_list[rd])]])
cur_flow.append(os.path.join(self.dir_,'flow',os.path.split(self.img_list[rd][rd2[i%len(self.img_list[rd])]])[1]))
cur_gt.append(self.label_list[rd][rd2[i%len(self.img_list[rd])]])
group_img=[]
group_flow=[]
group_gt=[]
for i in range(self.group):
tmp_img=imread(cur_img[i])
tmp_img=torch.from_numpy(img_normalize(tmp_img.astype(np.float32)/255.0))
tmp_img=F.interpolate(tmp_img.unsqueeze(0).permute(0,3,1,2),size=(self.img_size,self.img_size))
group_img.append(tmp_img)
tmp_gt=np.array(Image.open(cur_gt[i]).convert('L'))
tmp_gt=torch.from_numpy(tmp_gt.astype(np.float32)/255.0)
tmp_gt=F.interpolate(tmp_gt.view(1,tmp_gt.shape[0],tmp_gt.shape[1],1).permute(0,3,1,2),size=(self.img_size,self.img_size)).squeeze()
tmp_gt=tmp_gt.view(1,tmp_gt.shape[0],tmp_gt.shape[1])
group_gt.append(tmp_gt)
if self.use_flow==True:
tmp_flow=imread(cur_flow[i])
tmp_flow=torch.from_numpy(img_normalize(tmp_flow.astype(np.float32)/255.0))
tmp_flow=F.interpolate(tmp_flow.unsqueeze(0).permute(0,3,1,2),size=(self.img_size,self.img_size))
group_flow.append(tmp_flow)
group_img=(torch.cat(group_img,0))
if self.use_flow==True:
group_flow=torch.cat(group_flow,0)
group_gt=(torch.cat(group_gt,0))
if self.use_flow==True:
return group_img,group_flow,group_gt
else:
return group_img,group_gt