-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy pathtest_net.py
157 lines (122 loc) · 5.54 KB
/
test_net.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
import os
import argparse
import multiprocessing
import warnings
import copy
import numpy as np
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
from skimage.io import imread, imsave
from yacs.config import CfgNode
from tqdm import tqdm
from models.dual_hrnet import get_model
from xview2 import XView2Dataset
from utils import safe_mkdir
multiprocessing.set_start_method('spawn', True)
warnings.filterwarnings("ignore")
os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
parser = argparse.ArgumentParser()
parser.add_argument('--config_path', type=str, default="",
help='the location of the data folder')
parser.add_argument('--data_path', type=str, required=True, default="/mnt/Dataset/xView2/v2",
help='the location of the data folder')
parser.add_argument('--ckpt_path', type=str, required=True, default="",
help='Path to checkpoint')
parser.add_argument('--result_dir', type=str, required=True, default="",
help='Path to save result submit and compare iamges')
parser.add_argument('--is_train_data', action='store_true', dest='is_train_data',
help='')
parser.add_argument('--is_use_gpu', action='store_true', dest='is_use_gpu',
help='')
args = parser.parse_args()
class ModelWraper(nn.Module):
def __init__(self, model, is_use_gpu=False, is_split_loss=True):
super(ModelWraper, self).__init__()
self.is_use_gpu = is_use_gpu
self.is_split_loss = is_split_loss
if self.is_use_gpu:
self.model = model.cuda()
else:
self.model = model
def forward(self, inputs_pre, inputs_post):
inputs_pre = Variable(inputs_pre)
inputs_post = Variable(inputs_post)
if self.is_use_gpu:
inputs_pre = inputs_pre.cuda()
inputs_post = inputs_post.cuda()
pred_dict = self.model(inputs_pre, inputs_post)
loc = F.interpolate(pred_dict['loc'], size=inputs_pre.size()[2:4], mode='bilinear')
if self.is_split_loss:
cls = F.interpolate(pred_dict['cls'], size=inputs_post.size()[2:4], mode='bilinear')
else:
cls = None
return loc, cls
def argmax(loc, cls):
loc = torch.argmax(loc, dim=1, keepdim=False)
cls = torch.argmax(cls, dim=1, keepdim=False)
cls = cls + 1
cls[loc == 0] = 0
return loc, cls
def main():
if args.config_path:
with open(args.config_path, 'rb') as fp:
config = CfgNode.load_cfg(fp)
else:
config = None
ckpt_path = args.ckpt_path
result_submit_dir = os.path.join(args.result_dir, 'submit/')
result_compare_dir = os.path.join(args.result_dir, 'compare/')
dataset_mode = 'test' if not args.is_train_data else 'train'
imgs_dir = os.path.join(args.data_path, 'test/images/') if dataset_mode == 'test' \
else os.path.join(args.data_path, 'tier3/images/')
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.benchmark = True
print('data folder: ', args.data_path)
safe_mkdir(result_submit_dir)
safe_mkdir(result_compare_dir)
model = get_model(config)
model.load_state_dict(torch.load(ckpt_path, map_location='cpu')['state_dict'])
model.eval()
model_wrapper = ModelWraper(model, args.is_use_gpu, config.MODEL.IS_SPLIT_LOSS)
# model_wrapper = nn.DataParallel(model_wrapper)
model_wrapper.eval()
testset = XView2Dataset(args.data_path, rgb_bgr='rgb', preprocessing={'flip': False, 'scale': None, 'crop': None},
mode=dataset_mode)
testset_loader = torch.utils.data.DataLoader(testset, batch_size=1, shuffle=False, pin_memory=False, num_workers=1)
for i, samples in enumerate(tqdm(testset_loader)):
if dataset_mode == 'train' and i < 5520:
continue
inputs_pre = samples['pre_img']
inputs_post = samples['post_img']
image_ids = samples['image_id']
loc, cls = model_wrapper(inputs_pre, inputs_post)
if config.MODEL.IS_SPLIT_LOSS:
loc, cls = argmax(loc, cls)
loc = loc.detach().cpu().numpy().astype(np.uint8)
cls = cls.detach().cpu().numpy().astype(np.uint8)
else:
loc = torch.argmax(loc, dim=1, keepdim=False)
loc = loc.detach().cpu().numpy().astype(np.uint8)
cls = copy.deepcopy(loc)
for image_id, l, c in zip(image_ids, loc, cls):
localization_filename = 'test_localization_%s_prediction.png' % image_id
damage_filename = 'test_damage_%s_prediction.png' % image_id
imsave(os.path.join(result_submit_dir, localization_filename), l)
imsave(os.path.join(result_submit_dir, damage_filename), c)
pre_filename = 'test_pre_%s.png' % image_id
post_filename = 'test_post_%s.png' % image_id
pre_image = imread(os.path.join(imgs_dir, pre_filename))
post_image = imread(os.path.join(imgs_dir, post_filename))
mask_map_img = np.zeros((c.shape[0], c.shape[1], 3), dtype=np.uint8)
mask_map_img[c == 1] = (255, 255, 255)
mask_map_img[c == 2] = (229, 255, 50)
mask_map_img[c == 3] = (255, 159, 0)
mask_map_img[c == 4] = (255, 0, 0)
compare_img = np.concatenate((pre_image, mask_map_img, post_image), axis=1)
compare_filename = 'test_%s.png' % image_id
imsave(os.path.join(result_compare_dir, compare_filename), compare_img)
if __name__ == '__main__':
main()