-
Notifications
You must be signed in to change notification settings - Fork 3
/
inference.py
178 lines (143 loc) · 6.25 KB
/
inference.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
import argparse
from collections import defaultdict
import os
import os.path as osp
import torch
from models.tracker_manager import MotionTrackerManager
import tqdm
import copy
import yaml
import json
import pickle
from utils.coco_format import read_file
from utils.general_output import general_output
import utils.tracking_utils as tu
from datasets.video_parser import VID
cat_mapping = {
'kitti': {"car": 2, "pedestrian": 0, "cyclist": 1},
'nuscenes': {'bicycle': 0, 'motorcycle': 1, 'pedestrian': 2, 'bus': 3, 'car': 4, 'trailer': 5, 'truck': 6, 'construction_vehicle': 7, 'traffic_cone': 8, 'barrier': 9}
}
def parse_args():
parser = argparse.ArgumentParser(description="motion tracker")
parser.add_argument("--config", dest='config', help='settings of tracker in yaml format')
args = parser.parse_args()
return args
def single_gpu_test(
model,
dataset_name,
dets,
out_path: str = None,
img_infos: list = [],
):
outputs = defaultdict(list)
prog_bar = tqdm.tqdm(total=len(img_infos))
coco_outputs = defaultdict(list)
pred_id = 0
modelcats = cat_mapping[dataset_name]
for i, data in enumerate(img_infos):
with torch.no_grad():
img_info = copy.deepcopy(img_infos[i])
img_info.update({"img_info": img_infos[i]})
img_info.update({"calib": img_infos[i]["cali"]})
img_info.update({"frame_id": img_infos[i]["index"]})
if dataset_name == 'nuscenes': # for nuscenes
filename_clean = '/'.join(img_info['filename'].split('/')[2:])
det = dets[filename_clean]
#transform category id from mmdetection3d to ours
mmdet_cats = ['car', 'truck', 'trailer', 'bus', 'construction_vehicle', 'bicycle', 'motorcycle', 'pedestrian', 'traffic_cone', 'barrier']
boxes_3d = det['boxes_3d'][:,:7]
boxes_2d = det['boxes_2d']
scores_3d = det['scores_3d']
labels_3d = det['labels_3d']
dim = boxes_3d[:,3:6][:,[1,2,0]] #lhw(mmdet3D)->hwl(ours)
loc = boxes_3d[:,:3]
#TODO
loc[:,1] -= (dim[:,0] / 2.0)
dep = boxes_3d[:,2]
rot = boxes_3d[:,-1]
det_yaws = -1 * tu.alpha2yaw_torch(-1 *rot, loc[:, 0:1].view(-1), loc[:, 2:3].view(-1))
cat = [modelcats[mmdet_cats[i]] for i in labels_3d]
det_labels = torch.tensor(cat).view(-1).cuda()
det_bboxes = boxes_2d.cuda()
projection = det_bboxes.new_tensor(img_info['calib'])
det_depths = dep.cuda().view(-1,1).float()
det_dims = dim.cuda().float()
det_2dcs = tu.cameratoimage_torch(loc.cuda().float(), projection)
det_alphas = det_yaws.view(-1,1).cuda().float()
elif dataset_name == 'kitti': # for kitti
log_id = img_info["video_id"]
fr_id = img_info["frame_id"]
det = dets[log_id]["frames"][fr_id]["annotations"]
depth_results = []
bbox_results = []
dim_results = []
alpha_results = []
cen_2ds_results = []
depth_uncertainty_results = []
tracking_results = {}
label_results = []
for i, ann in enumerate(det):
depth_results.append(ann['depth'])
dim_results.append(ann['dimension'])
alpha_results.append(ann['alpha'])
cen_2ds_results.append(ann['box_center'])
bbox_results.append(ann['box']+[ann['score']])
label_results.append(modelcats[ann['obj_type'].lower()])
det_depths = torch.tensor(depth_results).cuda().view(-1,1)
det_bboxes = torch.tensor(bbox_results).cuda().view(-1,5)
det_labels = torch.tensor(label_results).cuda().view(-1)
det_dims = torch.tensor(dim_results).cuda().view(-1,3)
det_alphas = torch.tensor(alpha_results).cuda().view(-1,1)
det_2dcs = torch.tensor(cen_2ds_results).cuda().view(-1,2)
else:
raise NotImplementedError
det = {'det_labels': det_labels, 'det_bboxes': det_bboxes, 'det_depths': det_depths, 'det_dims': det_dims, 'det_2dcs': det_2dcs, 'det_alphas': det_alphas}
modelcats = cat_mapping[dataset_name]
result, use_3d_center = model.simple_test(
det_out=det, img_meta=img_info
)
coco_outputs, pred_id = general_output(
coco_outputs, result, img_info, use_3d_center, pred_id, modelcats, out_path
)
prog_bar.update()
prog_bar.close()
return coco_outputs
if __name__ == "__main__":
args = parse_args()
cfg = yaml.load(open(args.config, 'r'), Loader=yaml.Loader)
out_path = cfg['output_path']
info_path = cfg['info_path']
det_path = cfg['detection_path']
dataset_name = cfg['dataset_name']
if dataset_name == "kitti":
cats = [cat.capitalize() for cat in list(cat_mapping[dataset_name].keys())]
dets = read_file(det_path, category=cats)
elif dataset_name == "nuscenes":
with open(det_path, "rb") as f:
dets = pickle.load(f)['results']
else:
raise NotImplementedError
os.makedirs(out_path, exist_ok=True)
out_json_path = osp.join(out_path, "output.json")
img_infos = []
vid = VID(info_path)
vid_ids = vid.getVidIds()
for vid_id in vid_ids:
img_ids = vid.getImgIdsFromVidId(vid_id)
for img_id in img_ids:
info = vid.loadImgs([img_id])[0]
info["filename"] = info["file_name"]
info["type"] = "VID"
info["first_frame"] = True if info["index"] == 0 else False
img_infos.append(info)
model = MotionTrackerManager()
coco_outputs = single_gpu_test(
model,
dataset_name,
dets=dets,
out_path=out_path,
img_infos=img_infos,
)
print(f"\nwriting results to {out_path}")
with open(out_json_path, "w") as f:
json.dump(coco_outputs, f)