-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
325 lines (279 loc) · 16.2 KB
/
main.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
# -*-coding:utf-8-*
# author: wangxy
from __future__ import print_function
import os, numpy as np, time, cv2, torch
from os import listdir
from os.path import join
from file_operation.file import load_list_from_folder, mkdir_if_inexistence, fileparts
from detection.detection import Detection_2D, Detection_3D_only, Detection_3D_Fusion
from tracking.tracker import Tracker
from datasets.datafusion import datafusion2Dand3D
from datasets.coordinate_transformation import convert_3dbox_to_8corner, convert_x1y1x2y2_to_tlwh
from visualization.visualization_3d import show_image_with_boxes
os.environ["KMP_DUPLICATE_LIB_OK"]="TRUE"
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--data-path', type=str, default='datasets/kitti/train', help='dataset path')
parser.add_argument('--detections3D', type=str, default='3D_pointrcnn_Car_val', help='detections 3D foldername')
parser.add_argument('--detections2D', type=str, default='2D_rrc_Car_val', help='detections 2D foldername')
parser.add_argument('--save-path', type=str, default='results/train', help='result save path')
parser.add_argument('--save-img', action='store_true', help='save image')
parser.add_argument('--eval', action='store_true', help='start evaluation')
parser.add_argument('--gt-data', type=str, default=None, help='gt data dataframe path')
# parser.add_argument('--eval-output', type=str, default=None, help='eval result save path')
args = parser.parse_args()
# data_root = 'datasets/kitti/train'
# detections_name_3D = '3D_pointrcnn_Car_val'
# detections_name_2D = '2D_rrc_Car_val'
def is_image_file(filename):
return any(filename.endswith(extension) for extension in ['.png', '.jpg', '.jpeg', '.PNG', '.JPG', '.JPEG'])
def compute_color_for_id(label):
"""
Simple function that adds fixed color depending on the id
"""
palette = (2 ** 11 - 1, 2 ** 15 - 1, 2 ** 20 - 1)
color = [int((p * (label ** 2 - label + 1)) % 255) for p in palette]
return tuple(color)
class DeepFusion(object):
def __init__(self, max_age, min_hits):
'''
:param max_age: The maximum frames in which an object disappears.
:param min_hits: The minimum frames in which an object becomes a trajectory in succession.
'''
self.max_age = max_age
self.min_hits = min_hits
self.tracker = Tracker(max_age,min_hits)
self.reorder = [3, 4, 5, 6, 2, 1, 0]
self.reorder_back = [6, 5, 4, 0, 1, 2, 3]
self.frame_count = 0
def update(self,detection_3D_fusion,detection_2D_only,detection_3D_only,detection_3Dto2D_only,
additional_info, calib_file):
dets_3d_fusion = np.array(detection_3D_fusion['dets_3d_fusion'])
dets_3d_fusion_info = np.array(detection_3D_fusion['dets_3d_fusion_info'])
dets_3d_only = np.array(detection_3D_only['dets_3d_only'])
dets_3d_only_info = np.array(detection_3D_only['dets_3d_only_info'])
if len(dets_3d_fusion) == 0:
dets_3d_fusion = dets_3d_fusion
else:
dets_3d_fusion = dets_3d_fusion[:,self.reorder] # convert [h,w,l,x,y,z,rot_y] to [x,y,z,rot_y，l,w,h]
if len(dets_3d_only) == 0:
dets_3d_only = dets_3d_only
else:
dets_3d_only = dets_3d_only[:, self.reorder]
detection_3D_fusion = [Detection_3D_Fusion(det_fusion, dets_3d_fusion_info[i]) for i, det_fusion in enumerate(dets_3d_fusion)]
detection_3D_only = [Detection_3D_only(det_only, dets_3d_only_info[i]) for i, det_only in enumerate(dets_3d_only)]
detection_2D_only = [Detection_2D(det_fusion) for i, det_fusion in enumerate(detection_2D_only)]
self.tracker.predict_2d()
self.tracker.predict_3d()
self.tracker.update(detection_3D_fusion, detection_3D_only, detection_3Dto2D_only, detection_2D_only, calib_file, iou_threshold=0.5)
# self.tracker.update(detection_3D_fusion, detection_3D_only, detection_3Dto2D_only, detection_2D_only, calib_file, iou_threshold=0.01)
self.frame_count += 1
outputs = []
for track in self.tracker.tracks_3d:
if track.is_confirmed():
bbox = np.array(track.pose[self.reorder_back])
outputs.append(np.concatenate(([track.track_id_3d], bbox, track.additional_info)).reshape(1, -1))
if len(outputs) > 0:
outputs = np.stack(outputs, axis=0)
return outputs
@staticmethod
def _xywh_to_tlwh(bbox_xywh): # Convert the coordinate format of the bbox box from center x, y, w, h to upper left x, upper left y, w, h
if isinstance(bbox_xywh, np.ndarray):
bbox_tlwh = bbox_xywh.copy()
elif isinstance(bbox_xywh, torch.Tensor):
bbox_tlwh = bbox_xywh.clone()
bbox_tlwh[:, 0] = bbox_xywh[:, 0] - bbox_xywh[:, 2] / 2.
bbox_tlwh[:, 1] = bbox_xywh[:, 1] - bbox_xywh[:, 3] / 2.
return bbox_tlwh
def _tlwh_to_xyxy(self, bbox_tlwh):
x, y, w, h = bbox_tlwh
x1 = max(int(x), 0)
x2 = min(int(x+w), 0)
y1 = max(int(y), 0)
y2 = min(int(y+h), 0)
return x1, y1, x2, y2
def _tlwh_to_x1y1x2y2(self, bbox_tlwh):
x, y, w, h = bbox_tlwh
x1 = x
x2 = x + w
y1 = y
y2 = y + h
return x1, y1, x2, y2
if __name__ == '__main__':
# Define the file name
# data_root = 'datasets/kitti/train'
# detections_name_3D = '3D_pointrcnn_Car_val'
# detections_name_2D = '2D_rrc_Car_val'
data_root = args.data_path
detections_name_3D = args.detections3D
detections_name_2D = args.detections2D
# Define the file path
# calib_root = os.path.join(data_root, 'calib_train')
# dataset_dir = os.path.join(data_root,'image_02_train')
calib_root = os.path.join(data_root, 'calib')
dataset_dir = os.path.join(data_root,'image_02')
detections_root_3D = os.path.join(data_root, detections_name_3D)
detections_root_2D = os.path.join(data_root, detections_name_2D)
# Define the file path of results.
# save_root = 'results/train' # The root directory where the result is saved
save_root = args.save_path
txt_path_0 = os.path.join(save_root, 'data'); mkdir_if_inexistence(txt_path_0)
image_path_0 = os.path.join(save_root, 'image'); mkdir_if_inexistence(image_path_0)
# Open file to save in list.
# det_id2str = {1: 'Pedestrian', 2: 'Car', 3: 'Cyclist'}
# det_id2str = {1: 'Car', 2: 'Two_Wheeler', 3: 'Adult', 4: 'Kid', 5: 'SUV', 6: 'Van'}
det_id2str = {1: 'Car', 2: 'SUV_&_Van', 3: 'Truck', 4: 'Bus', 5: 'Special_Vehicle', 6: 'Two_Wheeler', 7: 'Person'}
# det_id2str = {1: 'Small_Car',
# 2: 'Light_Car',
# 3: 'Car',
# 4: 'Van',
# 5: 'SUV',
# 6: 'Small_Truck',
# 7: 'Medium_Truck',
# 8: 'Large_Truck',
# 9: 'Mini_Bus',
# 10: 'Bus',
# 11: 'Special_Vehicle',
# 12: 'Two_Wheeler',
# 13: 'Kickboard',
# 14: 'Adult',
# 15: 'Kid'}
calib_files = sorted(os.listdir(calib_root))
detections_files_3D = sorted(os.listdir(detections_root_3D))
detections_files_2D = sorted(os.listdir(detections_root_2D))
image_files = sorted(os.listdir(dataset_dir))
detection_file_list_3D, num_seq_3D = load_list_from_folder(detections_files_3D, detections_root_3D)
# print(detection_file_list_3D)
detection_file_list_2D, num_seq_2D = load_list_from_folder(detections_files_2D, detections_root_2D)
image_file_list, _ = load_list_from_folder(image_files, dataset_dir)
total_time, total_frames, i = 0.0, 0, 0 # Tracker runtime, total frames and Serial number of the dataset
tracker = DeepFusion(max_age=40, min_hits=1) # Tracker initialization
# Iterate through each data set
for seq_file_3D, image_filename in zip(detection_file_list_3D, image_files):
print('--------------Start processing the {} dataset--------------'.format(image_filename))
total_image = 0 # Record the total frames in this dataset
seq_file_2D = detection_file_list_2D[i]
seq_name, datasets_name, _ = fileparts(seq_file_3D)
txt_path = txt_path_0 + "/" + image_filename + '.txt'
image_path = image_path_0 + '/' + image_filename; mkdir_if_inexistence(image_path)
calib_file = [calib_file for calib_file in calib_files if calib_file==seq_name ]
calib_file_seq = os.path.join(calib_root, ''.join(calib_file))
image_dir = os.path.join(dataset_dir, image_filename)
image_filenames = sorted([join(image_dir, x) for x in listdir(image_dir) if is_image_file(x)])
seq_dets_3D = np.loadtxt(seq_file_3D, delimiter=',').reshape(-1, 15) # load 3D detections, N x 15
seq_dets_2D = np.loadtxt(seq_file_2D, delimiter=',').reshape(-1, 7) # load 2D detections, N x 6
# min_frame, max_frame = int(seq_dets_3D[:, 0].min()), len(image_filenames)
select_frames = list(set(seq_dets_3D[:, 0].astype(int).flatten().tolist()))
image_filenames = np.asarray(image_filenames)[select_frames]
# print(select_frames)
# print(np.asarray(image_filenames))
# print(image_filenames)
# for frame, img0_path in zip(range(min_frame, max_frame + 1), image_filenames):
for frame, img0_path in zip(select_frames, image_filenames):
# print(img0_path)
img_0 = cv2.imread(img0_path)
_, img0_name, _ = fileparts(img0_path)
dets_3D_camera = seq_dets_3D[seq_dets_3D[:, 0] == frame, 7:14] # 3D bounding box(h,w,l,x,y,z,theta)
dets_8corners = [convert_3dbox_to_8corner(det_tmp) for det_tmp in dets_3D_camera]
ori_array = seq_dets_3D[seq_dets_3D[:, 0] == frame, -1].reshape((-1, 1))
other_array = seq_dets_3D[seq_dets_3D[:, 0] == frame, 1:7]
additional_info = np.concatenate((ori_array, other_array), axis=1)
dets_3Dto2D_image = seq_dets_3D[seq_dets_3D[:, 0] == frame, 2:6]
# print(seq_dets_2D.shape)
# dets_2D = seq_dets_2D[seq_dets_2D[:, 0] == frame, 1:5] # 2D bounding box(x1,y1,x2,y2)
dets_2D = seq_dets_2D[seq_dets_2D[:, 0] == frame, 1:6] # 2D bounding box(class, x1,y1,x2,y2)
# Data Fusion(3D and 2D detections)
# detection_2D_fusion, detection_3Dto2D_fusion, detection_3D_fusion, detection_2D_only, detection_3Dto2D_only, detection_3D_only = \
# datafusion2Dand3D(dets_3D_camera, dets_2D, dets_3Dto2D_image, additional_info)
detection_2D_fusion, detection_3Dto2D_fusion, detection_3D_fusion, detection_2D_only, detection_3Dto2D_only, detection_3D_only, \
new_additional_info \
= datafusion2Dand3D(dets_3D_camera, dets_2D, dets_3Dto2D_image, additional_info)
detection_2D_only_tlwh = np.array([convert_x1y1x2y2_to_tlwh(i) for i in detection_2D_only]) # (x1,y1,x2,y2) to (x,y,center_x,center_y)
# print('detection_2D_fusion:', len(detection_2D_fusion))
# print('detection_3Dto2D_fusion:', len(detection_3Dto2D_fusion))
# print('detection_3D_fusion:', len(detection_3D_fusion['dets_3d_fusion']))
# print('detection_3Dto2D_only:', detection_3Dto2D_only.shape)
# print('detection_3D_only:', detection_3D_only['dets_3d_only'])
# print('detection_2D_only:', detection_2D_only.shape)
# print('-' * 50)
# just using fusion data
detection_3Dto2D_only = []
detection_3D_only = {'dets_3d_only': [], 'dets_3d_only_info': []}
start_time = time.time()
trackers = tracker.update(detection_3D_fusion, detection_2D_only_tlwh, detection_3D_only, detection_3Dto2D_only,
additional_info, calib_file_seq)
# trackers = tracker.update(detection_3D_fusion, detection_2D_only, detection_3D_only, detection_3Dto2D_only,
# new_additional_info, calib_file_seq)
cycle_time = time.time() - start_time
total_time += cycle_time
# Outputs
total_frames += 1 # Total frames for all datasets
total_image += 1 # Total frames for a dataset
if total_image % 50 == 0:
print("Now start processing the {} image of the {} dataset".format(total_image, image_filename))
if len(trackers) > 0:
for d in trackers:
bbox3d = d.flatten()
bbox3d_tmp = bbox3d[1:8] # 3D bounding box(h,w,l,x,y,z,theta)
id_tmp = int(bbox3d[0])
# id_tmp = int(bbox3d[0]) - int(trackers[0].flatten()[0]) # 라벨 번호 0부터 시작하기
ori_tmp = bbox3d[8]
type_tmp = det_id2str[bbox3d[9]]
bbox2d_tmp_trk = bbox3d[10:14]
conf_tmp = bbox3d[14]
color = compute_color_for_id(id_tmp)
# label = f'{id_tmp} {"car"}'
label = f'{id_tmp} {type_tmp}'
image_save_path = os.path.join(image_path, '%04d.jpg' % (int(img0_name)))
with open(txt_path, 'a') as f:
str_to_srite = '%d %d %s 0 0 %f %f %f %f %f %f %f %f %f %f %f %f %f\n' % (frame, id_tmp,type_tmp, ori_tmp,bbox2d_tmp_trk[0],
bbox2d_tmp_trk[1],bbox2d_tmp_trk[2],bbox2d_tmp_trk[3],bbox3d_tmp[0], bbox3d_tmp[1],bbox3d_tmp[2], bbox3d_tmp[3],
bbox3d_tmp[4], bbox3d_tmp[5],bbox3d_tmp[6],conf_tmp)
f.write(str_to_srite)
if args.save_img:
show_image_with_boxes(img_0, bbox3d_tmp, image_path, color, img0_name, label, calib_file_seq,line_thickness=1)
i += 1
print('--------------The time it takes to process all datasets are {}s --------------'.format(total_time))
print('--------------FPS = {} --------------'.format(total_frames/total_time))
# for evaluation
if args.eval:
import subprocess
import pandas as pd
from tqdm import tqdm
import re
import shutil
src = args.save_path
dst = args.save_path + '/trackeval'
os.makedirs(f'{dst}/gt/label_02', exist_ok=True)
os.makedirs(f'{dst}/trackers/label_02', exist_ok=True)
scenes = sorted(os.listdir(f'{src}/data'))
dp = pd.read_csv(args.gt_data, index_col=0, dtype={'frame':object})
# make gt, evaluate_tracking.seqmap.training
seqmap = [f'{scene.replace(".txt", "")} empty 0000 0010' for scene in scenes]
seqmap[0] = re.sub('0000', '0001', seqmap[0])
with open(f'{dst}/gt/evaluate_tracking.seqmap.training', 'w') as f:
f.write('\n'.join(seqmap))
for scene in tqdm(scenes):
scene = scene.replace('.txt', '')
# make gt, label_02
scene_df = dp.loc[dp['scene']==scene].copy()
scene_df[['truncated', 'occluded', 'alpha']]= 0
scene_df['frame'] = scene_df['frame'].astype(int)
scene_df[['frame']]
scene_df = scene_df[['frame', 'id', 'class', 'truncated', 'occluded', 'alpha', 'min_x', 'min_y', 'max_x', 'max_y', 'h', 'w', 'l', 'point_x', 'point_y', 'point_z', 'rot_y']]
dropped_duple_idx = scene_df[['frame', 'id']].drop_duplicates().index
scene_df = scene_df.loc[dropped_duple_idx].copy()
scene_df.to_csv(f'{dst}/gt/label_02/{scene}.txt', index=None, header=None, sep=' ')
# make trakers, label_02
if os.path.isfile(f'{src}/data/{scene}.txt')==False:
null_data = '0 0 None 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'
with open(f'{dst}/trackers/label_02/{scene}.txt', 'w') as f:
f.write(null_data)
else:
shutil.copy(f'{src}/data/{scene}.txt', f'{dst}/trackers/label_02/{scene}.txt')
os.chdir('./TrackEval_aivill')
GT_FOLDER = dst + '/gt'
TRACKERS_FOLDER = dst + '/trackers'
OUTPUT_FOLDER = dst
run_trackeval = f'python scripts/run_kitti.py --GT_FOLDER {GT_FOLDER} --TRACKERS_FOLDER {TRACKERS_FOLDER} --OUTPUT_FOLDER {OUTPUT_FOLDER}'
subprocess.call(run_trackeval, shell=True)