-
Notifications
You must be signed in to change notification settings - Fork 11
/
evaluate_model.py
232 lines (202 loc) · 7.99 KB
/
evaluate_model.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
import argparse
import logging
import os
import torch
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
matplotlib.use('Agg')
from loader import data_loader
from parser import get_evaluation_parser
from models import STGAT
from utils import *
def get_generator(checkpoint):
'''
Upload model
'''
n_units = (
[args.traj_lstm_hidden_size]
+ [int(x) for x in args.hidden_units.strip().split(",")]
+ [args.graph_lstm_hidden_size]
)
n_heads = [int(x) for x in args.heads.strip().split(",")]
model = STGAT(
obs_len=args.obs_len,
fut_len=args.fut_len,
n_coordinates=args.n_coordinates,
traj_lstm_hidden_size=args.traj_lstm_hidden_size,
n_units=n_units,
n_heads=n_heads,
graph_network_out_dims=args.graph_network_out_dims,
dropout=args.dropout,
alpha=args.alpha,
graph_lstm_hidden_size=args.graph_lstm_hidden_size,
noise_dim=args.noise_dim,
noise_type=args.noise_type,
add_confidence=args.add_confidence,
counter=args.counter,
)
model.load_state_dict(checkpoint["state_dict"])
model.cuda()
model.eval()
return model
def cal_ade_fde(fut_traj, pred_fut_traj):
'''
Compute the ADE and FDE
'''
ade = displacement_error(pred_fut_traj, fut_traj, mode="raw")
fde = final_displacement_error(pred_fut_traj[-1], fut_traj[-1], mode="raw")
return ade, fde
def evaluate(args, loader, generator):
'''
Evaluate the performances
'''
ade_outer, fde_outer = [], []
total_traj = 0
step = 0
with torch.no_grad():
for batch in loader:
batch = [tensor.cuda() for tensor in batch]
(
obs_traj,
fut_traj,
obs_traj_rel,
_,
seq_start_end,
) = batch
step += seq_start_end.shape[0]
ade, fde = [], []
total_traj += fut_traj.size(1)
for _ in range(args.best_k):
pred_fut_traj_rel = generator(
obs_traj_rel,
seq_start_end,
0, # No Teacher
3
)
pred_fut_traj = relative_to_abs(pred_fut_traj_rel, obs_traj[-1,:,:2])
ade_, fde_ = cal_ade_fde(fut_traj, pred_fut_traj)
ade.append(ade_)
fde.append(fde_)
ade_sum_batch = evaluate_helper(ade, seq_start_end)
fde_sum_batch = evaluate_helper(fde, seq_start_end)
ade_outer.append(ade_sum_batch)
fde_outer.append(fde_sum_batch)
ade_sum = sum(ade_outer)
fde_sum = sum(fde_outer)
return ade_sum, fde_sum, total_traj
def sceneplot(obsv_scene, pred_scene, gt_scene, figname='scene.png', lim=9.0):
'''
Plot a scene
'''
num_traj = pred_scene.shape[0]
obsv_frame = obsv_scene.shape[1]
pred_frame = pred_scene.shape[1]
cm_subsection = np.linspace(0.0, 1.0, num_traj)
colors = [matplotlib.cm.jet(x) for x in cm_subsection]
for i in range(num_traj):
for k in range(1, obsv_frame):
plt.plot(obsv_scene[i, k-1:k+1, 0], obsv_scene[i, k-1:k+1, 1],
'-o', color=colors[i], alpha=1.0)
plt.plot([obsv_scene[i, -1, 0], pred_scene[i, 0, 0]], [obsv_scene[i, -1, 1], pred_scene[i, 0, 1]],
'--', color=colors[i], alpha=1.0, linewidth=1.0)
for k in range(1, pred_frame):
alpha = 1.0 - k / pred_frame
width = (1.0 - alpha) * 24.0
plt.plot(pred_scene[i, k-1:k+1, 0], pred_scene[i, k-1:k+1, 1],
'--', color=colors[i], alpha=alpha, linewidth=width)
xc = obsv_scene[:, -1, 0].mean()
yc = obsv_scene[:, -1, 1].mean()
plt.xlim(xc-lim, xc+lim)
plt.ylim(yc-lim/2.0, yc+lim/2.0)
plt.gca().set_aspect('equal', adjustable='box')
plt.gca().get_xaxis().set_visible(False)
plt.gca().get_yaxis().set_visible(False)
plt.savefig(figname, bbox_inches='tight', pad_inches=.1)
plt.close()
def visualize(args, loader, generator):
'''
Viasualize some scenes
'''
keywords = args.resume.split('_')
suffix = 'ds_' + args.domain_shifts + '_' + keywords[1] + '_irm_' + keywords[3] + '.png'
# range of idx for visualization
lb_idx = 44
ub_idx = 44
with torch.no_grad():
for b, data in enumerate(loader):
batch = [tensor.cuda() for tensor in data]
(
obs_traj,
fut_traj,
obs_traj_rel,
_,
seq_start_end,
) = batch
for k in range(args.best_k):
pred_fut_traj_rel = generator(
obs_traj_rel,
seq_start_end,
0, # No Teacher
3
)
pred_fut_traj = relative_to_abs(pred_fut_traj_rel, obs_traj[-1,:,:2])
idx_sample = seq_start_end.shape[0]
for i in range(idx_sample):
if i < lb_idx or i > ub_idx:
continue # key scenes
idx_start, idx_end = seq_start_end[i][0], seq_start_end[i][1]
obsv_scene = obs_traj[:, idx_start:idx_end, :]
pred_scene = pred_fut_traj[:, idx_start:idx_end, :]
gt_scene = fut_traj[:, idx_start:idx_end, :]
figname = 'images/visualization/scene_{:02d}_{:02d}_sample_{:02d}_{}'.format(i, b, k, suffix)
sceneplot(obsv_scene.permute(1,0,2).cpu().detach().numpy(), pred_scene.permute(1,0,2).cpu().detach().numpy(), gt_scene.permute(1,0,2).cpu().detach().numpy(), figname)
def compute_col(predicted_traj, predicted_trajs_all, thres=0.2, num_interp=4):
'''
Compute the collisions
'''
dense_all = interpolate_traj(predicted_trajs_all, num_interp)
dense_ego = interpolate_traj(predicted_traj[None, :], num_interp)
distances = np.linalg.norm(dense_all - dense_ego, axis=-1)
mask = distances[:, 0] > 0
return distances[mask].min(axis=0) < thres
def main(args):
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu_num
checkpoint = torch.load(os.path.join('./models/', args.dataset_name, args.resume))
generator = get_generator(checkpoint)
path = get_dset_path(args.dataset_name, args.dset_type)
files = os.listdir(path)
envs_path = [os.path.join(path, file) for file in files]
loaders = [data_loader(args, env_path) for env_path in envs_path]
logging.info('Model: {}'.format(args.resume))
logging.info('Dataset: {}'.format(args.dataset_name))
logging.info('Eval shift: {}'.format(args.domain_shifts))
logging.info('Dataset type: {}'.format(args.dset_type))
# quantitative
if args.metrics == 'quantitative':
ade = 0
fde = 0
total_traj = 0
for loader in loaders:
ade_sum_i, fde_sum_i, total_traj_i = evaluate(args, loader, generator)
ade += ade_sum_i
fde += fde_sum_i
total_traj += total_traj_i
ade = ade / (total_traj * args.fut_len)
fde = fde / total_traj
logging.info('ADE: {:.4f}\tFDE: {:.4f}'.format(ade, fde))
# qualitative
if args.metrics == 'qualitative':
for loader in loaders:
visualize(args, loader, generator)
# collisions [to be implemented]
if args.metrics == 'collisions':
for loader in loaders:
visualize(args, loader, generator)
if __name__ == "__main__":
args = get_evaluation_parser().parse_args()
set_logger(os.path.join(args.log_dir, args.dataset_name, args.resume[:-8]+"_"+args.dset_type+"_ds_"+str(args.domain_shifts)+".log"))
torch.manual_seed(args.seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
main(args)