-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathtest2.py
executable file
·233 lines (186 loc) · 8.68 KB
/
test2.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
import argparse
import os
import numpy as np
from model.dataset.factory import get_imdb
from model.utils.config import cfg
from model.roi_data_layer.layer import RoIDataLayer
import torch
from model.SSH import SSH
from model.network import save_check_point, load_check_point
import cv2
from model.utils.timer import Timer
import torch.optim as optim
from model.utils.test_utils import _get_image_blob, _compute_scaling_factor, visusalize_detections
def parser():
parser = argparse.ArgumentParser('SSH Train module')
parser.add_argument('--gpu_ids', dest='gpu_ids', default='0', type=str,
help='gpu devices to be used')
parser.add_argument('--model_path', dest='model_path', default='check_point/check_point.zip', type=str,
help='Saved model path')
parser.add_argument('--model_save_path', dest='model_save_path', default='check_point/check_point.zip', type=str,
help='Saved model path')
parser.add_argument('--max_iters', dest='max_iters', default=450000, type=int,
help='maximum iterations')
return parser.parse_args()
def get_training_roidb(imdb):
"""
Get the training roidb given an imdb
:param imdb: The training imdb
:return: The training roidb
"""
def filter_roidb(roidb):
"""
Filtering samples without positive and negative training anchors
:param roidb: the training roidb
:return: the filtered roidb
"""
def is_valid(entry):
# Valid images have:
# (1) At least one foreground RoI OR
# (2) At least one background RoI
overlaps = entry['max_overlaps']
# find boxes with sufficient overlap
fg_inds = np.where(overlaps >= cfg.TRAIN.ANCHOR_POSITIVE_OVERLAP)[0]
# Select background RoIs as those within [BG_THRESH_LO, BG_THRESH_HI)
bg_inds = np.where((overlaps < cfg.TRAIN.BG_THRESH_HI) &
(overlaps >= cfg.TRAIN.BG_THRESH_LOW))[0]
# image is only valid if such boxes exist
valid = len(fg_inds) > 0 or len(bg_inds) > 0
return valid
num = len(roidb)
filtered_roidb = [entry for entry in roidb if is_valid(entry)]
num_after = len(filtered_roidb)
return filtered_roidb
# Augment imdb with flipped images
if cfg.TRAIN.USE_FLIPPED:
print('Appending horizontally-flipped training examples...')
imdb.append_flipped_images()
print('done')
print('Preparing training data...')
# Add required information to imdb
imdb.prepare_roidb()
# Filter the roidb
final_roidb = filter_roidb(imdb.roidb)
print('done')
return final_roidb
def train(net, optimizer, imdb, roidb, arg):
max_iters = arg.max_iters
iter = 1
display_interval = cfg.TRAIN.DISPLAY
train_data = RoIDataLayer(roidb, imdb.num_classes)
loss_sum = 0
m3_ssh_cls_loss_sum = 0
m3_bbox_loss_sum = 0
m2_ssh_cls_loss_sum = 0
m2_bbox_loss_sum = 0
m1_ssh_cls_loss_sum = 0
m1_bbox_loss_sum = 0
timer = {"forward": Timer(), "data": Timer()}
im = cv2.imread("/home/dwang/SynologyDrive/pyt_example/data/datasets/wider/WIDER_train/images/28--Sports_Fan/28_Sports_Fan_Sports_Fan_28_39.jpg")
im_scale = _compute_scaling_factor(im.shape, cfg.TRAIN.SCALES[0], cfg.TRAIN.MAX_SIZE)
bbox = [122, 2, 752, 688]
bbox = np.array([[bbox[0], bbox[1], bbox[0] + bbox[2], bbox[1] + bbox[3],1]], np.float)
bbox *=im_scale
bbox [:,4]=1
im_blob = _get_image_blob(im, [im_scale])[0]
im_info = np.array([[im_blob['data'].shape[2], im_blob['data'].shape[3], im_scale]])
im_data = im_blob['data']
im_data = torch.from_numpy(im_data).to(device)
# add a batch dimension
im_info = torch.from_numpy(im_info).to(device)
gt_boxes = torch.from_numpy(bbox).to(device).unsqueeze(0).float()
# img = np.squeeze(blobs['data'])
#
# img=img.transpose(1,2,0)
# for i in range(len(blobs['gt_boxes'])):
# pt1 = tuple(blobs['gt_boxes'][i, 0:2])
# pt2 = tuple(blobs['gt_boxes'][i, 2:4])
# cv2.rectangle(img, pt1, pt2, (255, 255, 255))
# cv2.imwrite("train.jpg", img)
optimizer.zero_grad()
m3_ssh_cls_loss, m2_ssh_cls_loss, m1_ssh_cls_loss, \
m3_bbox_loss, m2_bbox_loss, m1_bbox_loss = net(im_data, im_info, gt_boxes)
loss = (m3_ssh_cls_loss + m2_ssh_cls_loss + m1_ssh_cls_loss + \
m3_bbox_loss + m2_bbox_loss + m1_bbox_loss)
m3_ssh_cls_loss_sum += m3_ssh_cls_loss.item()
m3_bbox_loss_sum += m3_bbox_loss.item()
m2_ssh_cls_loss_sum += m2_ssh_cls_loss.item()
m2_bbox_loss_sum += m2_bbox_loss.item()
m1_ssh_cls_loss_sum += m1_ssh_cls_loss.item()
m1_bbox_loss_sum += m1_bbox_loss.item()
loss.backward()
torch.nn.utils.clip_grad_norm_(net.parameters(),0.5)
optimizer.step()
loss_sum += loss.item()
timer["forward"].toc()
# if m3_bbox_loss.item() == 0 :
# img = np.squeeze(blobs['data'])
#
# img=img.transpose(1,2,0)
# for i in range(len(blobs['gt_boxes'])):
# pt1 = tuple(blobs['gt_boxes'][i, 0:2])
# pt2 = tuple(blobs['gt_boxes'][i, 2:4])
# cv2.rectangle(img, pt1, pt2, (255, 255, 255))
# cv2.imwrite("zero/loss_0_{}.jpg".format(iter), img)
# f = open("zero/loss_0_{}.txt".format(iter), "a")
# f.write(blobs['file_path'])
if (iter % display_interval == 0):
loss_average = loss_sum / display_interval
m3_ssh_cls_loss_average = m3_ssh_cls_loss_sum / display_interval
m3_bbox_loss_average = m3_bbox_loss_sum / display_interval
m2_ssh_cls_loss_average = m2_ssh_cls_loss_sum / display_interval
m2_bbox_loss_average = m2_bbox_loss_sum / display_interval
m1_ssh_cls_loss_average = m1_ssh_cls_loss_sum / display_interval
m1_bbox_loss_average = m1_bbox_loss_sum / display_interval
loss_sum = 0
m3_ssh_cls_loss_sum = 0
m3_bbox_loss_sum = 0
m2_ssh_cls_loss_sum = 0
m2_bbox_loss_sum = 0
m1_ssh_cls_loss_sum = 0
m1_bbox_loss_sum = 0
print("------------------------iteration {}-----------{} left---------".format(iter, max_iters - iter))
print("Average per iter: {:.4f} second. ETA: {:.4f} hours".format(timer["forward"].average_time,
(max_iters - iter) * (
timer["forward"].average_time) / (
60 * 60)))
print("Average data load time: {:.4f}".format(timer["data"].average_time))
print('loss:{}\nm3 cls:{}\nm3 box:{}\nm2 cls:{}\nm2 box:{}'
'\nm1 cls:{}\nm1 box:{} '.format(loss_average, m3_ssh_cls_loss_average, m3_bbox_loss_average,
m2_ssh_cls_loss_average, m2_bbox_loss_average,
m1_ssh_cls_loss_average, m1_bbox_loss_average))
timer["forward"].reset()
timer["data"].reset()
if iter % cfg.TRAIN.CHECKPOINT == 0:
save_check_point(arg.model_save_path, iter, loss, net, optimizer)
print("check point saved")
if __name__ == '__main__':
arg = parser()
vgg16_image_net = True
if (os.path.isfile(arg.model_path)):
vgg16_image_net = False
imdb = get_imdb('wider_train')
roidb = get_training_roidb(imdb)
assert len(str(arg.gpu_ids)) == 1, "only single gpu is supported, " \
"use train_dist.py for multiple gpu support"
os.environ['CUDA_VISIBLE_DEVICES'] = str(arg.gpu_ids)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
net = SSH(vgg16_image_net)
optimizer = optim.SGD(net.parameters(), lr=cfg.TRAIN.LEARNING_RATE,
momentum=cfg.TRAIN.MOMENTUM, weight_decay=cfg.TRAIN.WEIGHT_DECAY)
if not vgg16_image_net:
check_point = load_check_point(arg.model_path)
net.load_state_dict(check_point['model_state_dict'])
# iter = check_point['iteration']
optimizer.load_state_dict(check_point['optimizer_state_dict'])
# for param_tensor in net.state_dict():
# print(param_tensor, "\t", net.state_dict()[param_tensor].size())
for state in optimizer.state.values():
for k, v in state.items():
if isinstance(v, torch.Tensor):
state[k] = v.to(device)
# for var_name in optimizer.state_dict():
# print(var_name, "\t", optimizer.state_dict()[var_name])
net.to(device)
net.train()
train(net, optimizer, imdb, roidb, arg)