-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathprecompute_sssdd.py
185 lines (178 loc) · 8.25 KB
/
precompute_sssdd.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
import math
import os
import random
import re
import cv2
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torch.utils.data
from torch.autograd import Variable
from torchvision import transforms
import imutils
import utils
from base_class import BaseModel, SegBaseModel, SSDDBaseModel, PascalDataset
import ssdd_function as ssddF
import time
from PIL import Image
from network import SegmentationPsa, PredictDiff, PredictDiffHead
import math
import cv2
cv2.setNumThreads(0)
############################################################
# dataset
############################################################
class SssddData(PascalDataset):
def __init__(self, dataset, config):
super().__init__(dataset, config)
self.label_dic = dataset.label_dic
self.joint_transform_list=[
None,
imutils.RandomHorizontalFlip(),
imutils.RandomResizeLong(448, 448),
imutils.RandomCrop(448),
None,
]
def __getitem__(self, image_index):
image_id = self.image_ids[image_index]
impath = self.config.VOC_ROOT+'/JPEGImages/'
imn = impath+image_id+'.jpg'
img = Image.open(imn).convert("RGB")
gt_class_mlabel = torch.from_numpy(self.label_dic[image_id])
gt_class_mlabel_bg = torch.from_numpy(np.concatenate(([1],self.label_dic[image_id])))
psan = 'prepare_labels/results/out_aff/'+image_id+'.npy'
psa=np.array(list(np.load(psan).item().values())).transpose(1,2,0)
psan = 'prepare_labels/results/out_aff_crf/'+image_id+'.npy'
psa_crf=np.load(psan).transpose(1,2,0)
h=psa.shape[0]
w=psa.shape[1]
img_norm, img_org, psa, psa_crf = self.img_label_resize([img, np.array(img), psa, psa_crf])
img_org = cv2.resize(img_org,self.config.OUT_SHAPE)
psa = cv2.resize(psa,self.config.OUT_SHAPE)
psa_crf = cv2.resize(psa_crf,self.config.OUT_SHAPE)
psa=self.get_prob_label(psa, gt_class_mlabel_bg).transpose(2,0,1)
psa_crf=self.get_prob_label(psa_crf, gt_class_mlabel_bg).transpose(2,0,1)
psa_mask = np.argmax(psa,0)
psa_crf_mask = np.argmax(psa_crf,0)
return img_norm, img_org, gt_class_mlabel, gt_class_mlabel_bg, psa_mask, psa_crf_mask
def __len__(self):
return self.image_ids.shape[0]
############################################################
# Models
############################################################
class SegModel(SegBaseModel):
def __init__(self, config):
super(SegModel, self).__init__(config)
self.config = config
in_channel=4096
self.seg_main = SegmentationPsa(config,num_classes=21, in_channel=in_channel, middle_channel=512, scale=2)
def set_bn_fix(m):
classname = m.__class__.__name__
if classname.find('BatchNorm') != -1:
for p in m.parameters(): p.requires_grad = False
self.apply(set_bn_fix)
def forward(self, inputs):
x, img_org, gt_class_mlabel = inputs
feats = self.encoder(x)
[x1,x2,x3,x4,x5] = feats
seg_outs_main = self.get_seg(self.seg_main, x5, gt_class_mlabel)
return seg_outs_main, feats
class SSDDModel(SSDDBaseModel):
def __init__(self, config):
super(SSDDModel, self).__init__(config)
self.dd_head0 = PredictDiffHead(config, in_channel=512, in_channel2=128)
self.dd0 = PredictDiff(config, in_channel=256, in_channel2=128)
def forward(self, inputs):
(seg_outs_main, feats), psa_mask, psa_crf_mask, gt_class_mlabel = inputs
[x1,x2,x3,x4,x5] = feats
x1=F.avg_pool2d(x1, 2, 2)
# first step
seg_main, seg_prob_main, seg_mask_main, seg_head_main = seg_outs_main
ignore_flags0=torch.from_numpy(ssddF.get_ignore_flags(psa_mask, psa_crf_mask, gt_class_mlabel)).cuda().float()
dd_head0 = self.dd_head0((seg_head_main.detach(), x1.detach()))
dd00 = ssddF.get_dd(self.dd0, dd_head0, psa_mask)
dd01 = ssddF.get_dd(self.dd0, dd_head0, psa_crf_mask)
dd_outs0 = ssddF.get_dd_mask(dd00, dd01, psa_mask, psa_crf_mask, ignore_flags0, dd_bias=0.1, bg_bias=0.1)
return dd_outs0
############################################################
# Precompute
############################################################
class Precompute():
def __init__(self, config, model_dir, model, weight_files):
super(Precompute, self).__init__()
self.config = config
self.model_dir = model_dir
self.epoch = 0
self.layer_regex = {
"lr1": r"(encoder.*)",
"lr10": r"(seg_main.*)",
"dd": r"(dd0.*)|(dd_head0.*)",
}
lr_1x = self.layer_regex["lr1"]
lr_10x = self.layer_regex["lr10"]
dd = self.layer_regex['dd']
seg_model=model[0].cuda()
ssdd_model=model[1].cuda()
self.param_lr_1x = [param for name, param in seg_model.named_parameters() if bool(re.fullmatch(lr_1x, name)) and not 'bn' in name]
self.param_lr_10x = [param for name, param in seg_model.named_parameters() if bool(re.fullmatch(lr_10x, name)) and not 'bn' in name]
self.param_dd = [param for name, param in ssdd_model.named_parameters() if bool(re.fullmatch(dd, name)) and not 'bn' in name]
lr=1e-3
self.seg_model=nn.DataParallel(seg_model)
self.ssdd_model=nn.DataParallel(ssdd_model)
self.seg_model.load_state_dict(torch.load(weight_files[0]))
self.ssdd_model.load_state_dict(torch.load(weight_files[1]))
def precompute_model(self, train_dataset):
# Data generators
self.train_set = SssddData(train_dataset, self.config)
train_generator = torch.utils.data.DataLoader(self.train_set, batch_size=self.config.BATCH, shuffle=False, num_workers=8, pin_memory=True)
self.seg_model.eval()
self.ssdd_model.eval()
self.cnt=0
for inputs in train_generator:
self.precompute_step(inputs)
def precompute_step(self, inputs):
img_norm, img_org, gt_class_mlabel, gt_class_mlabel_bg, psa_mask, psa_crf_mask = inputs
img_norm = Variable(img_norm).cuda().float()
img_org = Variable(img_org).cuda().float()
gt_class_mlabel = Variable(gt_class_mlabel).cuda().float()
gt_class_mlabel_bg = Variable(gt_class_mlabel_bg).cuda().float()
seg_outs = self.seg_model((img_norm, img_org, gt_class_mlabel_bg))
dd_outs = self.ssdd_model((seg_outs, psa_mask, psa_crf_mask, gt_class_mlabel))
seg_outs_main, feats = seg_outs
seg_main, seg_prob_main, seg_mask_main, _ = seg_outs_main
dd_outs0 = dd_outs
(dd00, dd01, ignore_flags0, refine_mask0) = dd_outs0
psa_mask = Variable(psa_mask).cuda().long()
psa_crf_mask = Variable(psa_crf_mask).cuda().long()
img_org=img_org.data.cpu().numpy()[...,::-1]
for i in range(len(img_norm)):
sid='_'+self.phase+'_'+self.saveid+'_'+str(self.cnt)
saven = self.savedir + '/D'+sid+'.png'
mask_png = utils.mask2png(saven, refine_mask0[i].squeeze().data.cpu().numpy())
saven = self.savedir + '/dk'+sid+'.png'
tmp=F.sigmoid(dd00)[i].squeeze().data.cpu().numpy()
cv2.imwrite(saven,tmp*255)
saven = self.savedir + '/da'+sid+'.png'
tmp=F.sigmoid(dd01)[i].squeeze().data.cpu().numpy()
cv2.imwrite(saven,tmp*255)
saven = self.savedir +'/dk'+sid
np.save(saven,dd00[i].data.cpu().numpy())
saven = self.savedir +'/da'+sid
np.save(saven,dd01[i].data.cpu().numpy())
print(self.cnt)
self.cnt += 1
def set_log_dir(self, phase, saveid, model_path=None):
self.phase = phase
self.saveid = saveid
self.savedir = 'precompute/'+self.saveid
print("save the results to "+self.savedir)
if not os.path.exists(self.savedir):
os.makedirs(self.savedir)
def models(config, weight_file=None):
seg_model = SegModel(config=config)
seg_model.initialize_weights()
ssdd_model = SSDDModel(config=config)
ssdd_model.initialize_weights()
return (seg_model, ssdd_model)