-
Notifications
You must be signed in to change notification settings - Fork 1
/
dataset.py
417 lines (375 loc) · 16.5 KB
/
dataset.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
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
import os
import math
import random
import torch.utils.data as data
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import os
from os import listdir
from os.path import isfile, join
import pandas as pd
from sklearn.model_selection import train_test_split
import gc;
gc.enable() # memory is tight
import torch
import pdb
import torch.utils.data as data
import cv2
from image import flip, color_aug
from image import get_affine_transform, affine_transform
dtype = "float32"
## HeatMap Genrating Functions
def gaussian_radius(det_size, min_overlap=0.7):
height, width = det_size
a1 = 1
b1 = (height + width)
c1 = width * height * (1 - min_overlap) / (1 + min_overlap)
sq1 = np.sqrt(b1 ** 2 - 4 * a1 * c1)
r1 = (b1 + sq1) / 2
a2 = 4
b2 = 2 * (height + width)
c2 = (1 - min_overlap) * width * height
sq2 = np.sqrt(b2 ** 2 - 4 * a2 * c2)
r2 = (b2 + sq2) / 2
a3 = 4 * min_overlap
b3 = -2 * min_overlap * (height + width)
c3 = (min_overlap - 1) * width * height
sq3 = np.sqrt(b3 ** 2 - 4 * a3 * c3)
r3 = (b3 + sq3) / 2
return min(r1, r2, r3)
def gaussian2D(shape, sigma=1):
m, n = [(ss - 1.) / 2. for ss in shape]
y, x = np.ogrid[-m:m + 1, -n:n + 1]
h = np.exp(-(x * x + y * y) / (2 * sigma * sigma))
h[h < np.finfo(h.dtype).eps * h.max()] = 0
return h
def gaussian1D(shape, sigma=1):
m = (shape - 1.) / 2.
y = np.ogrid[-m:m + 1]
h = np.exp(-(y * y) / (2 * sigma))
h[h < np.finfo(h.dtype).eps * h.max()] = 0
return h
def draw_umich_gaussian(heatmap, center,y_0, y_1, x_0, x_1, radius, k=1, is_train_x0 = True):
diameter = 2 * radius + 1
#pdb.set_trace()
#gaussian = gaussian1D((diameter), sigma=diameter / 6)
gaussian = gaussian2D((diameter, diameter), sigma=diameter / 6) # array([[0.01831564, 0.13533528, 0.01831564],
# [0.13533528, 1. , 0.13533528],
# [0.01831564, 0.13533528, 0.01831564]])
y, x= int(center[1]), int(center[0])
height, width = heatmap.shape[0:2]
left, right = min(x, radius), min(width - x, radius + 1)
top, bottom = min(y, radius), min(height - y, radius + 1)
masked_gaussian = gaussian[radius:radius + bottom, radius] # [0:3, 0:3]
masked_gaussian_low = gaussian[radius - top:radius + 1 , radius] # [0:3, 0:3]
if is_train_x0:
#masked_heatmap = heatmap[y - top :y + bottom, x_0 : x_1]
#masked_gaussian = np.expand_dims(gaussian[radius - top:radius + bottom,radius],axis=1) * np.ones_like(masked_heatmap)
masked_heatmap = heatmap[y_0:y_0 + bottom, x_0] # array([[0., 0., 0.],
# [0., 0., 0.],
# [0., 0., 0.]], dtype=float32)
masked_heatmap_e = heatmap[y_0:y_0 + bottom, x_1]
if min(masked_gaussian.shape) > 0 and (min(masked_heatmap.shape)> 0 or min(masked_heatmap_e.shape)> 0) : # TODO debug
np.maximum(masked_heatmap, masked_gaussian * k, out=masked_heatmap)
np.maximum(masked_heatmap_e, masked_gaussian * k, out=masked_heatmap_e)
else:
masked_heatmap_low = heatmap[y_1 - top:y_1 + 1, x_0]
masked_heatmap_lowe = heatmap[y_1 - top:y_1 + 1, x_1]
if min(masked_gaussian.shape) > 0 and (
min(masked_heatmap_low.shape) > 0 or min(masked_heatmap_lowe.shape) > 0): # TODO debug
np.maximum(masked_heatmap_low, masked_gaussian_low * k, out=masked_heatmap_low)
np.maximum(masked_heatmap_lowe, masked_gaussian_low * k, out=masked_heatmap_lowe)
return heatmap
def draw_umich_ind(ind_heatmap, center,x_0, x_1):
y, x= int(center[1]), int(center[0])
ind_heatmap[y, x_0 : x_1] = 1
return ind_heatmap
def reverse_res_to_bbox(res):
down_ratio = res['input'].shape[1] / res['hm'].shape[1]
output_h, output_w = res['hm'].shape[0], res['hm'].shape[1]
num_objs = np.sum(res['reg_mask'])
bbox = np.zeros((num_objs, 4), dtype='float32')
ct = res['reg'][:num_objs]
ct[:, 0] += res['ind'][:num_objs] % output_w
ct[:, 1] += res['ind'][:num_objs] // output_w
h, w = res['wh'][:num_objs, 1], res['wh'][:num_objs, 0]
bbox[:, 0] = (ct[:, 0] * 2 - w) / 2
bbox[:, 2] = (ct[:, 0] * 2 + w) / 2
bbox[:, 1] = (ct[:, 1] * 2 - h) / 2
bbox[:, 3] = (ct[:, 1] * 2 + h) / 2
bbox *= down_ratio
return bbox
def reverse_res_to_bbox_centerline_ind(res):
#pdb.set_trace()
down_ratio = res['input'].shape[1] / res['hm'].shape[1]
output_h, output_w = res['hm'].shape[0], res['hm'].shape[1]
num_objs = np.sum(res['reg_mask'])
bbox = np.zeros((num_objs, 4), dtype='float32')
ct = np.zeros((num_objs, 2), dtype="float32")
ct_y = res['reg_y'][:num_objs]
bbox[:, 0] = res['ind'][:num_objs][:,0] % output_w
bbox[:, 2] = res['ind'][:num_objs][:,1] % output_w
ct_y += res['ind'][:num_objs][:,0] // output_w
ct[:, 1] = ct_y
#h, w = res['wh'][:num_objs, 1], res['wh'][:num_objs, 0]
h_d, h_u = res['h_ud'][:num_objs, 1], res['h_ud'][:num_objs, 0]
bbox[:, 1] = (ct[:, 1] + h_d)
bbox[:, 3] = (ct[:, 1] + h_u)
bbox *= down_ratio
return bbox
def reverse_res_to_bbox_centerline(res):
#pdb.set_trace()
down_ratio = res['input'].shape[1] / res['hm'].shape[1]
output_h, output_w = res['hm'].shape[0], res['hm'].shape[1]
num_objs = np.sum(res['reg_mask'])
bbox = np.zeros((num_objs, 4), dtype='float32')
ct = np.zeros((num_objs, 2), dtype="float32")
ct_y = np.squeeze(res['reg_y'][:num_objs])
bbox[:, 0] = res['ind'][:num_objs][:,0] % output_w
bbox[:, 2] = res['ind'][:num_objs][:,1] % output_w
ct_y += res['ind'][:num_objs][:,0] // output_w
#ct[:, 1] = ct_y
#h, w = res['wh'][:num_objs, 1], res['wh'][:num_objs, 0]
h_ud = np.squeeze(res['h_ud'][:num_objs])
bbox[:, 1] = ct_y
bbox[:, 3] = (ct_y + h_ud)
bbox *= down_ratio
return bbox
class ctDataset(data.Dataset):
def __init__(self, split="train"):
# img_dir = 'kitti dataset/'
# ../input/kitti_single/training/label_2/
self.keep_res = False
self.split = split
self.not_rand_crop = True
self.no_color_aug = False
self.original_shift = 0
self.original_scale = 0
self.scale = [0, 0.3, 0.5, 0.7]#, 1.2, 1.5, 1.8, 2.0]
self.shift = [0, 0.1, 0.2, 0.3]#, 0.5, 0.7, 0.8, 0.6]
# self.scale = [0, 0.7, 0.9, 1.2]
# self.shift = [0, 0.1,0.2]
self.flip = 0.5
self.scale_shift_values = 0.4
self.num_classes = 4
self.max_objs = 100
self.down_ratio = 2
self.color_aug_value = 0.2
self.input_h, self.input_w = 512, 512
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.mean = [0.64566372, 0.64566372, 0.64566372]
self.std = [0.08311639, 0.08311639, 0.08311639]
root = '/media/zjn/F/Data/TX_data/all_bbox/'
self.root = root
self.image_dir = os.path.join(root, "RGB/4K_800_801_data_imag_add_mode/")
self.label_dir = os.path.join(root, "label_txt/4K_800_801_label_txt_add_mode_only_start-stop-points/")
if self.split == "train":
imageset_txt = os.path.join(root, "imagesets_new", "train.txt")
elif self.split == "test":
imageset_txt = os.path.join(root, "imagesets_new", "test.txt")
elif self.split == "validation":
imageset_txt = os.path.join(root, "imagesets_new", "validation.txt")
image_files = []
for line in open(imageset_txt, "r"):
image_name = line.replace("\n", "")
image_files.append(image_name)
# image_name_2 = image_name.replace("RGB_500","RGB_1000")
# image_files.append(image_name_2)
# image_name_3 = image_name.replace("RGB_500", "RGB_2000")
# image_files.append(image_name_3)
# image_name_4 = image_name.replace("RGB_500", "RGB_4000")
# image_files.append(image_name_4)
self.image_files = image_files
self.label_files = [i.replace(".jpg", ".txt") for i in self.image_files]
self.num_samples = len(self.image_files)
def _coco_box_to_bbox(self, box):
bbox = np.array([box[1], box[0], box[3], box[2]],
dtype=np.float32)
return bbox
def _get_border(self, border, size):
i = 1
while size - border // i <= border // i:
i *= 2
return border // i
def __len__(self):
return self.num_samples
def __getitem__(self, idx):
# pdb.set_trace()
original_idx = self.image_files[idx].replace(".jpg", "")
img_path = os.path.join(self.image_dir + self.image_files[idx])
img = cv2.imread(img_path)
# default_resolution = [1601, 400] # [375, 1242]
# height, width = img.shape[0], img.shape[1]
# print("Before pre-reshape: ", height, width)
# transform to 512 * 512
# reshape image to default size (some samples have slightly different size)
height, width = img.shape[0], img.shape[1]
cc = np.array([img.shape[1] / 2., img.shape[0] / 2.], dtype=np.float32)
if self.keep_res:
input_h = (height | self.pad) + 1
input_w = (width | self.pad) + 1
ss = np.array([input_w, input_h], dtype=np.float32)
else:
ss = max(img.shape[0], img.shape[1]) * 1.0
input_h, input_w = self.input_h, self.input_w
# pdb.set_trace()
flipped = False
if self.split == "train":
if not self.not_rand_crop:
ss = ss * np.random.choice(np.arange(0.6, 1.4, 0.1))
w_border = self._get_border(128, img.shape[1])
h_border = self._get_border(128, img.shape[0])
cc[0] = np.random.randint(low=w_border, high=img.shape[1] - w_border)
cc[1] = np.random.randint(low=h_border, high=img.shape[0] - h_border)
else:
if np.random.random() < self.scale_shift_values:
sf = np.random.choice(self.scale)
cf = np.random.choice(self.shift)
else:
sf = self.original_scale
cf = self.original_shift
cc[0] += ss * np.clip(np.random.randn() * cf, -2 * cf, 2 * cf)
cc[1] += ss * np.clip(np.random.randn() * cf, -2 * cf, 2 * cf)
ss = ss * np.clip(np.random.randn() * sf + 1, 1 - sf, 1 + sf)
if np.random.random() < self.flip:
# pdb.set_trace()
flipped = True
img = img[::-1, ::-1, :]
cc[0] = width - cc[0] - 1
# pdb.set_trace()
trans_input = get_affine_transform(cc, ss, 0, [input_w, input_h])
# mapping = np.array([[1, 0, 0], [0, 1, 0]]).astype(np.float)
inp = cv2.warpAffine(img, trans_input, (input_w, input_h))
# print(inp)
# inp = cv2.resize(img, (input_w, input_h))
# scale_h, scale_w = input_h / height, input_w / width
inp = (inp.astype(np.float32) / 255.)
if np.random.random() < self.color_aug_value:
color_aug(self._data_rng, inp, self._eig_val, self._eig_vec)
# inp = (inp - self.mean) / self.std
inp = inp.transpose(2, 0, 1)
inp = inp.astype(np.float32)
output_h = input_h // self.down_ratio
output_w = input_w // self.down_ratio
num_classes = self.num_classes
trans_output = get_affine_transform(cc, ss, 0, [output_w, output_h])
img = cv2.resize(img, (width, height))
if self.split in ["test","validation"]:
res = {'image': img, \
'input': inp, \
'index': idx, \
'ori_index': original_idx}
return res
# get the labels
label_path = os.path.join(self.label_dir + self.label_files[idx])
with open(label_path) as f:
content = f.readlines()
content = [x.split() for x in content]
# print(content)
draw_gaussian = draw_umich_gaussian
draw_ind = draw_umich_ind
hm = np.zeros((num_classes, output_h, output_w), dtype=np.float32)
reg_mask = np.zeros((self.max_objs), dtype=np.uint8)
cls = np.zeros((self.max_objs), dtype=np.float32)
# con_signal = np.zeros((max_objs), dtype=np.int16)
#wh = np.zeros((self.max_objs, 2), dtype=np.float32)
h_ud = np.zeros((self.max_objs, 1), dtype=np.float32)
reg = np.zeros((self.max_objs, 2), dtype=np.float32)
reg_y = np.zeros((self.max_objs, 1), dtype=np.float32)
ind = np.zeros((self.max_objs, 2), dtype=np.int64)
#ind_hm = np.zeros((num_classes, output_h, output_w), dtype=np.float32)
count = 0
for c in content:
# if (c[0] == "Car"):
#pdb.set_trace()
bbox = np.array(c[1:5], dtype="float32")
bbox = self._coco_box_to_bbox(bbox)
if flipped:
bbox[[0, 2]] = width - bbox[[2, 0]] - 1
bbox[[1, 3]] = height - bbox[[3, 1]] - 1
# bbox_new3 = width - bbox[1] - 1
# bbox[1] = width - bbox[3] - 1
# bbox[3] = bbox_new3
# bbox_new2 = height - bbox[0] - 1
# bbox[0] = height - bbox[2] - 1
# bbox[2] = bbox_new2
#if int(c[0]) == 1:
cls_id = int(c[5])
# print("zhijie1:",cls_id)
#else:
#cls_id = int(c[5]) + 3
# print("zhijie1:",cls_id)
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)
ct_int = ct.astype(np.int32)
x_0 = bbox[0].astype(np.int32)
x_1 = bbox[2].astype(np.int32)
y_0 = bbox[1].astype(np.int32)
y_1 = bbox[3].astype(np.int32)
draw_gaussian(hm[cls_id], ct_int, y_0, y_1, x_0, x_1, radius, is_train_x0 = True)
#pdb.set_trace()
h_u = bbox[3] - bbox[1]
#wh[count] = 1. * h_u, 1. * h
h_ud[count] = 1. * h_u
ind_w = np.array([y_0 * output_w + bbox[0], y_0 * output_w + bbox[2]], dtype=np.float32)
ind[count] = ind_w
#draw_ind(ind_hm[cls_id], ct_int, x_0, x_1)
reg[count] = ct - ct_int
reg_y[count] = bbox[1] - y_0
reg_mask[count] = 1
cls[count] = cls_id
count = count + 1
res = {'image': img, \
'input': inp, \
'hm': hm, \
'reg_mask': reg_mask, \
'ind': ind, \
#'ind_hm': ind_hm, \
#'wh': wh, \
'h_ud': h_ud, \
#'reg': reg, \
'reg_y': reg_y, \
'cl': cls, \
# 'con_id':con_signal, \
'index': idx, \
'ori_index': original_idx}
return res
if __name__ == "__main__":
im_idx = 33
my_dataset = ctDataset()
res = my_dataset.__getitem__(im_idx)
#pdb.set_trace()
img = res['image']
inp = res['input'].transpose(1, 2, 0)
hm = res['hm']
plt.title("Original Image")
plt.imshow(img)
plt.show()
plt.title("transform Image")
plt.imshow(inp)
plt.show()
plt.title("Ground Truth Heat Map")
im = plt.imshow(hm[0])
plt.colorbar(im)
plt.savefig('./Hm-0.png', dpi=1000, bbox_inches='tight') # 保存曲线图
plt.show()
plt.title("Ground Truth Heat Map")
bbox = reverse_res_to_bbox_centerline(res)
for b in bbox:
cv2.rectangle(inp, (int(b[0]), int(b[1])), (int(b[2]), int(b[3])), (0, 0, 1), 2)
plt.title("Calculated bounding box positions")
plt.imshow(inp)
plt.show()