-
Notifications
You must be signed in to change notification settings - Fork 3
/
train_infer_A2GNN_others.py
181 lines (142 loc) · 8.62 KB
/
train_infer_A2GNN_others.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
import torch
import torchvision
from tool import imutils
import argparse
import importlib
import numpy as np
from pygcn.A2GNN import A2GNN
import voc12.data
from torch.utils.data import DataLoader
import torch.nn.functional as F
import os.path
import torch.optim as optim
import cv2
from model_loss_semseg_gatedcrf import ModelLossSemsegGatedCRF
import imageio
import mytool
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument("--weights", default='./netWeights/final_model/aff_scribble.pth', type=str)
parser.add_argument("--network", default="network.resnet38_aff_gated", type=str)
parser.add_argument("--infer_list", default="voc12/train_aug.txt", type=str)
parser.add_argument("--num_workers", default=8, type=int)
parser.add_argument("--voc12_root", default='/your/path/VOCdevkit/VOC2012', type=str)
parser.add_argument("--cam_dir", default='/your/path/cam', type=str)
parser.add_argument("--seed_label_root", default='./data/Init_Label/Scribble_SuperPixel', type=str)
parser.add_argument('--radius', type=int, default=3)
parser.add_argument('--epochs', type=int, default=50)
parser.add_argument("--alpha", default=6, type=float)
parser.add_argument("--rw_weight", default=3, type=int)
parser.add_argument("--logt", default=6, type=int)
parser.add_argument("--beta", default=8, type=int)
parser.add_argument('--hidden', type=int, default=256, help='Number of hidden units.')
parser.add_argument('--dropout', type=float, default=0.5, help='Dropout rate (1 - keep probability).')
parser.add_argument('--lr', type=float, default=0.03, help='Initial learning rate.')
parser.add_argument('--weight_decay', type=float, default=5e-4, help='Weight decay (L2 loss on parameters).')
parser.add_argument("--num_class", default=21, type=int)
parser.add_argument("--save_path", default="./out/scribble_pred", type=str)
parser.add_argument("--rw", default=False, type=bool)
args = parser.parse_args()
model = getattr(importlib.import_module(args.network), 'Net')()
model.load_state_dict(torch.load(args.weights), strict=False)
model.eval()
model.cuda()
infer_dataset = voc12.data.VOC12ImageDataset(args.infer_list, voc12_root=args.voc12_root,
transform=torchvision.transforms.Compose(
[np.asarray,
model.normalize,
imutils.HWC_to_CHW]))
infer_data_loader = DataLoader(infer_dataset, shuffle=False, num_workers=args.num_workers, pin_memory=True)
for iter, (name, img) in enumerate(infer_data_loader):
name = name[0]
if os.path.exists(os.path.join(args.save_path, name+'.png')):
print(iter, 'has finished')
else:
print(iter)
img_nopad = img.clone()
orig_shape = img.shape
padded_size = (int(np.ceil(img.shape[2]/8)*8), int(np.ceil(img.shape[3]/8)*8))
p2d = (0, padded_size[1] - img.shape[3], 0, padded_size[0] - img.shape[2])
img = F.pad(img, p2d)
dheight = int(np.ceil(img.shape[2]/8))
dwidth = int(np.ceil(img.shape[3]/8))
with torch.no_grad():
features, aff_mat, _ = model.forward(img.cuda(), radius=args.radius, to_dense=False)
aff_mat = torch.pow(aff_mat, 1).squeeze(dim=0)
f_h, f_w = features.shape[-2], features.shape[-1]
aff_mat = mytool.generate_aff(f_h, f_w, aff_mat, radius=args.radius)
aff_cropping = mytool.generate_aff_cropping(f_h, f_w, radius=args.radius)
normalized_img = F.interpolate(img, [f_h, f_w], mode='bilinear', align_corners=False)
normalized_input = {'rgb': normalized_img.cuda()}
seed_label = cv2.imread(os.path.join(args.seed_label_root,name+'.png'),cv2.IMREAD_GRAYSCALE)
gt_h, gt_w = seed_label.shape[-2], seed_label.shape[-1]
seed_label = np.pad(seed_label, ((0, p2d[3]), (0, p2d[1])), mode='constant')
seed_label = torch.from_numpy(seed_label).unsqueeze(dim=0)
seed_label = seed_label.long()
adj = aff_mat.cuda()
adj = adj - torch.diag(torch.diag(adj))
adj = adj + torch.eye(adj.shape[0]).cuda()
features = features.squeeze(dim=0)
features = features.permute(1,2,0)
features = features.view(-1,5632)
features = features/torch.sum(features, dim=1, keepdim=True)
gnn_model = A2GNN(nfeat=features.shape[1],
nhid=args.hidden,
nclass=args.num_class,
nlayers=3,
dropout_rate=args.dropout)
gatedcrf = ModelLossSemsegGatedCRF()
critersion = torch.nn.CrossEntropyLoss(weight=None, ignore_index=255, reduction='elementwise_mean').cuda()
optimizer = optim.Adam(gnn_model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
gnn_model.cuda()
optimizer.zero_grad()
for epoch in range(args.epochs):
loss_train, gnn_model = mytool.train_with_nocrf(epoch, gnn_model, features, adj, seed_label, critersion,
f_h, f_w, aff_cropping, optimizer)
for epoch in range(args.epochs):
loss_train, gnn_model = mytool.train_with_crf(epoch, gnn_model, features, adj, seed_label, critersion,
f_h, f_w, gatedcrf, normalized_input, aff_cropping, optimizer)
pred = mytool.infer_nocrf(gnn_model, features, adj, f_h, f_w, aff_cropping)
pred = F.interpolate(torch.unsqueeze(pred, dim=0), [gt_h, gt_w], mode='bilinear', align_corners=False)
#------------------------------------RW-----------------------------------------------------
if args.rw==True:
aff_mat = torch.pow(aff_mat, args.beta)
aff_mat = torch.squeeze(aff_mat)
trans_mat = aff_mat / torch.sum(aff_mat, dim=0, keepdim=True)
for _ in range(args.logt):
trans_mat = torch.matmul(trans_mat, trans_mat)
cam = np.load(os.path.join(args.cam_dir, name + '.npy'), allow_pickle=True).item()
cam_full_arr = np.zeros((21, orig_shape[2], orig_shape[3]), np.float32)
for k, v in cam.items():
cam_full_arr[k + 1] = v
cam_full_arr[0] = (1 - np.max(cam_full_arr[1:], (0), keepdims=False)) ** args.alpha
cam_full_arr = np.pad(cam_full_arr, ((0, 0), (0, p2d[3]), (0, p2d[1])), mode='constant')
cam_full_arr = torch.from_numpy(cam_full_arr)
cam_full_arr = F.avg_pool2d(cam_full_arr, 8, 8)
cam_vec = cam_full_arr.view(args.num_class, -1)
cam_rw = torch.matmul(cam_vec.cuda(), trans_mat.cuda())
cam_rw = cam_rw.view(1, args.num_class, dheight, dwidth)
cam_rw = F.interpolate(cam_rw, [img.shape[2], img.shape[3]], mode='bilinear',
align_corners=False)
cam_rw_pred = (cam_rw[:,:,:orig_shape[2], :orig_shape[3]]).cpu().data.numpy()
cam_rw_pred = np.squeeze(cam_rw_pred)
pred_probs = F.softmax(pred, dim=1).cpu().data.numpy()
pred_probs = np.squeeze(pred_probs)
original_img = np.array(imageio.imread(os.path.join(args.voc12_root + '/JPEGImages', name + '.jpg'))).astype(
np.uint8)
cam_pred = mytool.crf_inference_inf(original_img, pred_probs, labels=args.num_class)
cam_img = np.argmax(cam_pred, 0)
pred_probs = pred_probs + args.rw_weight*cam_rw_pred
pred_probs = pred_probs / np.sum(pred_probs, axis=0, keepdims=True)
crf_pred = mytool.crf_inference_inf(original_img, pred_probs, labels=args.num_class)
crf_pred = np.argmax(crf_pred, 0)
imageio.imsave(os.path.join(args.save_path, name + '.png'), crf_pred.astype(np.uint8))
else:
pred_probs = F.softmax(pred, dim=1).cpu().data.numpy()
pred_probs = np.squeeze(pred_probs)
original_img = np.array(imageio.imread(os.path.join(args.voc12_root + '/JPEGImages', name + '.jpg'))).astype(
np.uint8)
crf_pred = mytool.crf_inference_inf(original_img, pred_probs, labels=args.num_class)
crf_pred = np.argmax(crf_pred, 0)
imageio.imsave(os.path.join(args.save_path, name + '.png'), crf_pred.astype(np.uint8))
torch.cuda.empty_cache()