-
Notifications
You must be signed in to change notification settings - Fork 1
/
ctdet.py
223 lines (193 loc) · 8.93 KB
/
ctdet.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
import os
import cv2
import numpy as np
import pandas as pd
import math
import random
import torch.utils.data as data
from sklearn.externals import joblib
from image import flip, color_aug
from image import get_affine_transform, affine_transform
from image import gaussian_radius, draw_umich_gaussian, draw_msra_gaussian
class CTDetDataset(data.Dataset):
default_resolution = [512, 512]
mean = np.array([0.485, 0.456, 0.406],
dtype=np.float32).reshape(1, 1, 3)
std = np.array([0.229, 0.224, 0.225],
dtype=np.float32).reshape(1, 1, 3)
def _get_border(self, border, size):
i = 1
while size - border // i <= border // i:
i *= 2
return border // i
def __init__(self, opt, split):
super(CTDetDataset, self).__init__()
self.num_classes = opt.num_class
self.max_objs = 20
self.voc_color = [(v // 32 * 64 + 64, (v // 8) % 4 * 64, v % 8 * 32) \
for v in range(1, self.num_classes + 1)]
self._data_rng = np.random.RandomState(123)
self._eig_val = np.array([0.2141788, 0.01817699, 0.00341571],
dtype=np.float32)
self._eig_vec = np.array([
[-0.58752847, -0.69563484, 0.41340352],
[-0.5832747, 0.00994535, -0.81221408],
[-0.56089297, 0.71832671, 0.41158938]
], dtype=np.float32)
self.opt = opt
self.split = split
if split == 'train':
self.file_list = opt.train
elif split == 'val':
self.file_list = opt.val
else:
ValueError('not valid split!')
random_forest_model = 'random_forest.model'
self.random_forest = joblib.load(random_forest_model)
self.kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
def get_mask(self, image, anns):
shape = image.shape[:2]
mask = np.zeros(shape, dtype=np.uint8)
for ann in anns:
ann = [i if i >= 0 else 0 for i in ann]
patch = image[ann[1]: ann[3], ann[0]: ann[2]]
patch_shape = patch.shape
patch = patch.reshape((-1, 3))
df = pd.DataFrame({'B': patch[:, 0], 'G': patch[:, 1], 'R': patch[:, 2]})
pred = self.random_forest.predict_proba(df)
pred = pred[:, 1]
pred[pred < 0.5] = 0
pred = (pred * 255).astype(np.uint8)
pred = pred.reshape((patch_shape[0], patch_shape[1]))
pred = cv2.erode(pred, self.kernel)
pred = cv2.dilate(pred, self.kernel)
mask[ann[1]: ann[3], ann[0]: ann[2]] = pred
return mask
def __getitem__(self, index):
'''
img_info['image_path']： 序列帧的目录
img_info['label'] ： 序列帧对应的标签
'''
img_info = self.file_list[index]
img_path = img_info['image_path']
tick = len(img_path) // 2 # 取2帧做序列检测
start_num = random.randint(1, tick) - 1
img_1_path = img_path[start_num]
img_2_path = img_path[start_num + tick]
anns_all = img_info['label']
anns = anns_all[start_num]
num_objs = min(len(anns), self.max_objs)
if 'Pos' in img_1_path:
cls = 1
else:
cls = 0
image_1 = cv2.imread(img_1_path)
# image_2 = cv2.imread(img_2_path)
image_1_float = cv2.imread(img_1_path, 0).astype(np.float32)
image_2_float = cv2.imread(img_2_path, 0).astype(np.float32)
image_2 = abs(image_2_float - image_1_float)
image_2 = image_2 / np.max(image_2)
shape = image_1.shape[:2]
mask = np.zeros(shape, dtype=np.uint8)
if 'Pos' in img_1_path:
mask = self.get_mask(image_1, anns)
height, width = image_1.shape[0], image_1.shape[1]
c = np.array([image_1.shape[1] / 2., image_1.shape[0] / 2.], dtype=np.float32)
s = max(image_1.shape[0], image_1.shape[1]) * 1.0
input_h, input_w = self.default_resolution[0], self.default_resolution[1]
flipped = False
if self.split == 'train':
s = s * np.random.choice(np.arange(0.6, 1., 0.1))
w_border = self._get_border(500, image_1.shape[1])
h_border = self._get_border(300, image_1.shape[0])
c[0] = np.random.randint(low=w_border, high=image_1.shape[1] - w_border)
c[1] = np.random.randint(low=h_border, high=image_1.shape[0] - h_border)
if np.random.random() < self.opt.flip:
flipped = True
image_1 = image_1[:, ::-1, :]
image_2 = image_2[:, ::-1]
mask = mask[:, ::-1]
c[0] = width - c[0] - 1
trans_input = get_affine_transform(
c, s, 0, [input_w, input_h])
inp_1 = cv2.warpAffine(image_1, trans_input,
(input_w, input_h),
flags=cv2.INTER_LINEAR)
inp_2 = cv2.warpAffine(image_2, trans_input,
(input_w, input_h),
flags=cv2.INTER_LINEAR)
mask = cv2.warpAffine(mask, trans_input,
(input_w, input_h),
flags=cv2.INTER_LINEAR)
mask = cv2.resize(mask, (input_w//4, input_h//4))
inp_2 = cv2.resize(inp_2, (input_w//4, input_h//4))
# debug
# src_img = inp_1.copy()
inp_1 = (inp_1.astype(np.float32) / 255.)
# inp_2 = (inp_2.astype(np.float32) / 255.)
mask = (mask.astype(np.float32) / 255.)
if self.split == 'train':
color_aug(self._data_rng, inp_1, self._eig_val, self._eig_vec)
# color_aug(self._data_rng, inp_2, self._eig_val, self._eig_vec)
inp_1 = (inp_1 - self.mean) / self.std
# inp_2 = (inp_2 - self.mean) / self.std
inp_1 = inp_1.transpose(2, 0, 1)
# inp_2 = inp_2.transpose(2, 0, 1)
output_h = input_h // self.opt.down_ratio
output_w = input_w // self.opt.down_ratio
num_classes = self.num_classes
trans_output = get_affine_transform(c, s, 0, [output_w, output_h])
hm = np.zeros((num_classes, output_h, output_w), dtype=np.float32)
wh = np.zeros((self.max_objs, 2), dtype=np.float32)
reg = np.zeros((self.max_objs, 2), dtype=np.float32)
ind = np.zeros((self.max_objs), dtype=np.int64)
reg_mask = np.zeros((self.max_objs), dtype=np.uint8)
draw_gaussian = draw_umich_gaussian
pos = []
for k in range(num_objs):
bbox = np.array(anns[k][:4], dtype=np.float32)
cls_id = anns[k][-1]
if flipped:
bbox[[0, 2]] = width - bbox[[2, 0]] - 1
bbox[:2] = affine_transform(bbox[:2], trans_output)
bbox[2:] = affine_transform(bbox[2:], trans_output)
bbox[[0, 2]] = np.clip(bbox[[0, 2]], 0, output_w - 1)
bbox[[1, 3]] = np.clip(bbox[[1, 3]], 0, output_h - 1)
h, w = bbox[3] - bbox[1], bbox[2] - bbox[0]
if h > 0 and w > 0:
radius = gaussian_radius((math.ceil(h), math.ceil(w)))
radius = max(0, int(radius))
ct = np.array(
[(bbox[0] + bbox[2]) / 2, (bbox[1] + bbox[3]) / 2], dtype=np.float32)
pos.append([bbox[0], bbox[1], bbox[2], bbox[3]])
ct_int = ct.astype(np.int32)
draw_gaussian(hm[cls_id], ct_int, radius)
wh[k] = 1. * w, 1. * h
ind[k] = ct_int[1] * output_w + ct_int[0]
reg[k] = ct - ct_int
reg_mask[k] = 1
inp_2 = np.expand_dims(inp_2, axis=0)
mask = np.expand_dims(mask, axis=0) * 0.9 + hm * 0.1 # best (0.9 & 0.1: 85.93)
mask = mask / (np.max(mask) + 1e-9)
# mask = cv2.GaussianBlur(mask, (5, 5), 0)
ret = {'input1': inp_1, 'input2': inp_2, 'hm': hm, 'reg_mask': reg_mask, 'ind': ind, 'wh': wh,
'reg': reg, 'cls': cls, 'mask': mask}
# debug
'''
if num_objs:
src_img = cv2.resize(src_img, (128, 128))
heatmap = mask[0].copy()
heatmap = cv2.applyColorMap(np.uint8(255*heatmap), cv2.COLORMAP_JET)
heatmap = np.float32(heatmap) / np.max(heatmap)
cam = heatmap + np.float32(src_img/255)
cam = cam / np.max(cam)
cam = np.uint8(255 * cam)
# cam = cv2.GaussianBlur(cam, (5, 5), 0)
for item in pos:
cam = cv2.rectangle(cam, (item[0], item[1]), (item[2], item[3]), (255, 0, 0), 1)
base_name = os.path.basename(img_1_path)
cv2.imwrite(os.path.join('vis_output', base_name), cam)
'''
return ret
def __len__(self):
return len(self.file_list)