forked from brjathu/PHALP
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_tracker.py
267 lines (221 loc) · 16.5 KB
/
test_tracker.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision.utils import save_image, make_grid
import os
import traceback
import warnings
import json
import copy
import argparse
import joblib
import cv2
import time
import numpy as np
from tqdm import tqdm
from yacs.config import CfgNode as CN
from utils.make_video import render_frame_main
from utils.utils import str2bool, get_colors
from deep_sort_ import nn_matching
from deep_sort_.detection import Detection
from deep_sort_.tracker import Tracker
from PHALP import PHALP_tracker
from evaluate_PHALP import evaluate_trackers
warnings.filterwarnings("ignore")
RGB_tuples = get_colors()
def test_tracker(opt, phalp_tracker, checkpoint=None):
phalp_tracker.eval()
video_seq = np.load(opt.video_seq) if("npy" in opt.video_seq) else [opt.video_seq]
try:
os.system("mkdir out/" + opt.storage_folder)
os.system("mkdir out/" + opt.storage_folder + "/results")
except: pass
for video_file_name in tqdm(video_seq):
try:
if(opt.use_gt): track = joblib.load('_DATA/embeddings/v1_gt_' + str(video_file_name) + '.pickle')
else: track = joblib.load('_DATA/embeddings/v1_' + str(video_file_name) + '.pickle')
final_results_dic = {}
final_visuals_dic = {}
sequence = track.keys()
for video in sequence:
os.system("rm -rf out/" + opt.storage_folder + "/_TEMP/*")
metric = nn_matching.NearestNeighborDistanceMetric(opt, opt.hungarian_th, opt.past_lookback)
tracker = Tracker(opt, metric, max_age=opt.max_age_track, n_init=opt.n_init, phalp_tracker=phalp_tracker, dims=[4096, 4096, 99])
frame_list = sorted(list(track[video].keys()))
frame_list = frame_list[:]
frame_length = len(frame_list)
max_ids = opt.max_ids
person_id = torch.zeros(frame_length, max_ids) + -1
center = torch.zeros(frame_length, max_ids, 2)
scale = torch.zeros(frame_length, max_ids, 2)
bbox = torch.zeros(frame_length, max_ids, 4)
conf_c = torch.zeros(frame_length, max_ids, 1)
pose_emb = torch.zeros(frame_length, max_ids, 4096)
appe_emb = torch.zeros(frame_length, max_ids, 4096)
loca_emb = torch.zeros(frame_length, max_ids, 99) + 1
uv_vector = torch.zeros(frame_length, max_ids, 4, 256, 256)
frame_size = torch.zeros(frame_length, max_ids, 2)
image_name = np.zeros((frame_length, max_ids, 1), dtype=object)
mask_name = np.zeros((frame_length, max_ids, 1), dtype=object)
ground_truth = np.zeros((frame_length, max_ids, 1))
shots = np.zeros((frame_length,))
for frame_idx, frame in enumerate(frame_list):
idx = 0
for idx_ in range(len(track[video][frame])):
person_ = track[video][frame][idx_+1]
if(person_['score']>opt.low_th_c ):
person_id[frame_idx, idx] = 0
pose_emb[frame_idx, idx, 2048:] = torch.from_numpy(person_['pose_embedding'])
appe_emb[frame_idx, idx, :] = torch.from_numpy(person_['appe_embedding'])
loca_emb[frame_idx, idx, :93] = torch.from_numpy(person_['loca_embedding'])
conf_c[frame_idx, idx, :] = torch.from_numpy(np.array(person_['score']))
bbox[frame_idx, idx, :] = torch.from_numpy(np.array([person_['bbox'][0], person_['bbox'][1], person_['bbox'][0]+person_['bbox'][2], person_['bbox'][1]+person_['bbox'][3]]))
center[frame_idx, idx, :] = torch.from_numpy(np.array(person_['center']))
scale[frame_idx, idx, :] = torch.from_numpy(np.array(person_['scale']))
uv_vector[frame_idx, idx, :, :, :] = torch.from_numpy(person_['uv_vector'])
frame_size[frame_idx, idx, :] = torch.from_numpy(np.array(person_['image_size']))
mask_name[frame_idx, idx, :] = np.array([person_['mask_name']])
image_name[frame_idx, idx, :] = np.array([person_['image_name']])
ground_truth[frame_idx, idx, :] = np.array([person_['gt']])
try: shots[frame_idx] = person_['shot']
except: shots[frame_idx] = 0
idx += 1
if(idx>=opt.max_ids): break
BS, T, P = 1, frame_length, max_ids
window = frame_length//opt.window
embeddings = torch.cat((appe_emb, pose_emb, loca_emb), 2)
opt.shot = 0
tracked_frames = []
for w_ in range(frame_length//window):
for t in range(window):
if(opt.verbose): print("t : ", t, video_file_name, opt.storage_folder)
t_ = t+w_*window
loc_ = np.where(person_id[w_*window:(w_+1)*window][t]!=-1)[0]
detections = []; tracked_appe_single = []; tracked_time = []; tracked_mask_ = [];
tracked_ids = []; tracked_bbox = []; tracked_ids_ = [];
tracked_appe = []; tracked_pose = []; tracked_bbox_= [];
tracked_loca = []; tracked_pose_single = []; tracked_feat_= [];
for m in range(len(loc_)):
bbox_candidate = bbox[w_*window:(w_+1)*window][t][loc_][m]
w = bbox_candidate[2] - bbox_candidate[0]
h = bbox_candidate[3] - bbox_candidate[1]
opt.shot = shots[t_-1] if(t_>0) else 0
if(opt.track_dataset=="mupots"): th_h = 200
else: th_h = 100
if(h>th_h and w>50 ):
detection_data = {
"bbox" : [bbox_candidate[0], bbox_candidate[1], w, h],
"conf_c" : conf_c[w_*window:(w_+1)*window][t][loc_][m],
"embedding" : embeddings[w_*window:(w_+1)*window][t][loc_][m],
"uv_vector" : uv_vector[w_*window:(w_+1)*window][t][loc_][m],
"center" : center[w_*window:(w_+1)*window][t][loc_][m],
"scale" : scale[w_*window:(w_+1)*window][t][loc_][m],
"size" : frame_size[w_*window:(w_+1)*window][t][loc_][m],
"img_name" : image_name[w_*window:(w_+1)*window][t][loc_][m],
"mask_name" : mask_name[w_*window:(w_+1)*window][t][loc_][m],
"ground_truth" : ground_truth[w_*window:(w_+1)*window][t][loc_][m],
"time" : t_,
}
detections.append(Detection(detection_data))
tracker.predict()
matches = tracker.update(detections, t_, frame_list[t_], opt.shot)
for tracks_ in tracker.tracks:
if(frame_list[t_] not in tracked_frames): tracked_frames.append(frame_list[t_])
if(not(tracks_.is_confirmed())): continue
track_id = tracks_.track_id
bbox_ = tracks_.phalp_bbox[-1]
tracked_ids.append(track_id)
tracked_bbox.append([bbox_[0], bbox_[1], bbox_[2], bbox_[3]])
if(opt.render):
tracked_appe_single.append(tracks_.phalp_uv_map); tracked_appe.append(tracks_.phalp_uv_predicted); # phalp_uv_predicted
tracked_pose_single.append(tracks_.phalp_pose_features[-1]); tracked_pose.append(tracks_.phalp_pose_predicted);
tracked_loca.append(tracks_.phalp_loca_predicted); tracked_time.append(tracks_.time_since_update);
if(tracks_.time_since_update==0):
tracked_ids_.append(track_id)
tracked_bbox_.append([bbox_[0], bbox_[1], bbox_[2], bbox_[3]])
tracked_mask_.append(tracks_.detection_data[-1]['mask_name'])
if(tracks_.hits==opt.n_init):
for ia in range(opt.n_init-1):
try:
final_results_dic[tracked_frames[-2-ia]][0].append(track_id)
final_results_dic[tracked_frames[-2-ia]][1].append(tracks_.phalp_bbox[-2-ia])
if(opt.render):
final_visuals_dic[tracked_frames[-2-ia]][0].append(track_id)
final_visuals_dic[tracked_frames[-2-ia]][1].append(tracks_.phalp_bbox[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][3].append(tracks_.phalp_loca_predicted_[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][4][0].append(tracks_.phalp_pose_predicted_[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][4][1].append(tracks_.phalp_pose_features[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][5][0].append(tracks_.phalp_uv_map_[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][5][1].append(tracks_.phalp_uv_predicted_[-2-ia])
final_visuals_dic[tracked_frames[-2-ia]][6].append(0)
except:
final_results_dic.setdefault(tracked_frames[-2-ia], [[track_id], tracks_.phalp_bbox[-2-ia], t_-ia-1])
if(opt.render):
final_visuals_dic.setdefault(tracked_frames[-2-ia],
[[track_id], tracks_.phalp_bbox[-2-ia], t_-ia-1, tracks_.phalp_loca_predicted_[-2-ia], [tracks_.phalp_pose_predicted_[-2-ia],
tracks_.phalp_pose_features[-2-ia]], [tracks_.phalp_uv_map_[-2-ia], tracks_.phalp_uv_predicted_[-2-ia]], [0]])
if(frame_list[t_] in final_results_dic.keys() or frame_list[t_] in final_visuals_dic.keys()): print("Error!")
else:
final_results_dic.setdefault(frame_list[t_], [tracked_ids_, tracked_bbox_, t_])
if(opt.render): final_visuals_dic.setdefault(frame_list[t_], [tracked_ids, tracked_bbox, t_, tracked_loca, [tracked_pose, tracked_pose_single], [tracked_appe, tracked_appe_single], tracked_time])
frame_tracked = final_results_dic.keys()
for t_, frame_ in enumerate(frame_list):
if(frame_ not in frame_tracked): final_results_dic.setdefault(frame_, [[], [[]], t_])
save_loc = video_file_name.split("/")[0] + "______" + video_file_name.split("/")[1] if("AVA" in opt.track_dataset) else video_file_name
joblib.dump(final_results_dic, "out/" + opt.storage_folder + "/results/" + save_loc + ".pkl")
save_loc = video_file_name.split("/")[0] + "______" + video_file_name.split("/")[1] if("AVA" in opt.track_dataset) else video_file_name+"_distance"
joblib.dump(tracker.tracked_cost, "out/" + opt.storage_folder + "/results/" + save_loc + ".pkl")
if(opt.render):
t_ = 0
for frame_key in tqdm(final_visuals_dic.keys()):
rendered_, f_size = render_frame_main(opt, phalp_tracker, final_visuals_dic[frame_key][2], opt.track_dataset, [opt.base_path, opt.mask_path], video,
frame_key, final_visuals_dic[frame_key][0], final_visuals_dic[frame_key][1], final_visuals_dic[frame_key][3],
final_visuals_dic[frame_key][4], final_visuals_dic[frame_key][5], final_visuals_dic[frame_key][6],
number_of_windows=4, downsample=opt.downsample, storage_folder="out/" + opt.storage_folder + "/_TEMP/", track_id=-100)
if(t_==0):
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
video_loc = video.split("/")[0] + "______" + video.split("/")[1] if("AVA" in opt.track_dataset) else video
video_file = cv2.VideoWriter("out/" + opt.storage_folder + "/PHALP_" + opt.track_dataset + "_" + video + ".mp4", fourcc, 15, frameSize=f_size)
video_file.write(rendered_)
t_ += 1
video_file.release()
except Exception as e:
print(e)
print(traceback.format_exc())
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='PHALP_pixel Tracker')
parser.add_argument('--batch_id', type=int, default='-1')
parser.add_argument('--track_dataset', type=str, default='posetrack')
parser.add_argument('--predict', type=str, default='APL')
parser.add_argument('--storage_folder', type=str, default='Videos_v20.000')
parser.add_argument('--distance_type', type=str, default='A5')
parser.add_argument('--use_gt', type=str2bool, nargs='?', const=True, default=False)
parser.add_argument('--alpha', type=float, default=0.1)
parser.add_argument('--low_th_c', type=float, default=0.9)
parser.add_argument('--hungarian_th', type=float, default=100.0)
parser.add_argument('--track_history', type=int, default=7)
parser.add_argument('--max_age_track', type=int, default=20)
parser.add_argument('--n_init', type=int, default=5)
parser.add_argument('--max_ids', type=int, default=50)
parser.add_argument('--window', type=int, default=1)
parser.add_argument('--verbose', type=str2bool, nargs='?', const=True, default=False)
parser.add_argument('--base_path', type=str)
parser.add_argument('--mask_path', type=str)
parser.add_argument('--video_seq', type=str, default='_DATA/posetrack/list_videos_val.npy')
parser.add_argument('--version', type=str, default='v1')
parser.add_argument('--render', type=str2bool, nargs='?', const=True, default=False)
parser.add_argument('--render_type', type=str, default='HUMAN_HEAD_FAST')
parser.add_argument('--render_up_scale', type=int, default=2)
parser.add_argument('--res', type=int, default=256)
parser.add_argument('--downsample', type=int, default=1)
parser.add_argument('--encode_type', type=str, default='3c')
parser.add_argument('--cva_type', type=str, default='least_square')
parser.add_argument('--past_lookback', type=int, default=1)
opt = parser.parse_args()
phalp_tracker = PHALP_tracker(opt)
phalp_tracker.cuda()
phalp_tracker.eval()
phalp_tracker.HMAR.reset_nmr(512)
test_tracker(opt, phalp_tracker)