-
Notifications
You must be signed in to change notification settings - Fork 17
/
evalpc.py
211 lines (181 loc) · 10.1 KB
/
evalpc.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
import torch
import time
from datetime import datetime
from pathlib import Path
import numpy as np
from data.data_loader import get_dataloader, get_datasets
from utils.config import cfg
from utils.model_sl import load_model
from utils.hungarian import hungarian
from utils.loss_func import PermLoss
from utils.evaluation_metric import matching_accuracy, calcorrespondpc, summarize_metrics, print_metrics, compute_metrics, compute_transform
from parallel import DataParallel
from models.correspondSlover import SVDslover, RANSACSVDslover
from collections import defaultdict
def eval_model(model, dataloader, eval_epoch=None, metric_is_save=False, estimate_iters=1,
viz=None, usepgm=True, userefine=False, save_filetime='time'):
print('-----------------Start evaluation-----------------')
lap_solver = hungarian
permevalLoss = PermLoss()
since = time.time()
all_val_metrics_np = defaultdict(list)
iter_num = 0
dataset_size = len(dataloader.dataset)
print('train datasize: {}'.format(dataset_size))
device = next(model.parameters()).device
print('model on device: {}'.format(device))
if eval_epoch is not None:
if eval_epoch == 0:
model_path = str(Path(cfg.OUTPUT_PATH) / 'params' / 'params_best.pt')
print('Loading best model parameters')
load_model(model, model_path)
else:
model_path = str(Path(cfg.OUTPUT_PATH) / 'params' / 'params_{:04}.pt'.format(eval_epoch))
print('Loading model parameters from {}'.format(model_path))
load_model(model, model_path)
was_training = model.training
model.eval()
running_since = time.time()
for inputs in dataloader:
P1_gt, P2_gt = [_.cuda() for _ in inputs['Ps']]
n1_gt, n2_gt = [_.cuda() for _ in inputs['ns']]
# e1_gt, e2_gt = [_.cuda() for _ in inputs['es']]
A1_gt, A2_gt = [_.cuda() for _ in inputs['As']] # edge连接性
perm_mat = inputs['gt_perm_mat'].cuda()
T1_gt, T2_gt = [_.cuda() for _ in inputs['Ts']]
Inlier_src_gt, Inlier_ref_gt = [_.cuda() for _ in inputs['Ins']]
Label = torch.tensor([_ for _ in inputs['label']])
batch_cur_size = perm_mat.size(0)
iter_num = iter_num + 1
infer_time = time.time()
with torch.set_grad_enabled(False):
if cfg.EVAL.ITERATION:
P1_gt_copy = P1_gt.clone()
P2_gt_copy = P2_gt.clone()
P1_gt_copy_inv = P1_gt.clone()
P2_gt_copy_inv = P2_gt.clone()
s_perm_mat = caliters_perm(model, P1_gt_copy, P2_gt_copy, A1_gt, A2_gt, n1_gt, n2_gt, estimate_iters)
if cfg.EVAL.CYCLE:
s_perm_mat_inv = caliters_perm(model, P2_gt_copy_inv, P1_gt_copy_inv, A2_gt, A1_gt, n2_gt, n1_gt, estimate_iters)
s_perm_mat = s_perm_mat * s_perm_mat_inv.permute(0, 2, 1)
permevalloss=torch.tensor([0])
else:
s_prem_tensor, Inlier_src_pre, Inlier_ref_pre_tensor = model(P1_gt, P2_gt, A1_gt, A2_gt, n1_gt, n2_gt)
if cfg.EVAL.CYCLE:
s_prem_tensor_inv, Inlier_src_pre_inv, Inlier_ref_pre_tensor_inv = model(P2_gt, P1_gt, A2_gt, A1_gt, n2_gt, n1_gt)
if cfg.PGM.USEINLIERRATE:
s_prem_tensor = Inlier_src_pre * s_prem_tensor * Inlier_ref_pre_tensor.transpose(2, 1).contiguous()
if cfg.EVAL.CYCLE:
s_prem_tensor_inv = Inlier_src_pre_inv * s_prem_tensor_inv * \
Inlier_ref_pre_tensor_inv.transpose(2,1).contiguous()
permevalloss = permevalLoss(s_prem_tensor, perm_mat, n1_gt, n2_gt)
s_perm_mat = lap_solver(s_prem_tensor, n1_gt, n2_gt, Inlier_src_pre, Inlier_ref_pre_tensor)
if cfg.EVAL.CYCLE:
s_perm_mat_inv = lap_solver(s_prem_tensor_inv, n2_gt, n1_gt, Inlier_src_pre_inv, Inlier_ref_pre_tensor_inv)
s_perm_mat = s_perm_mat*s_perm_mat_inv.permute(0,2,1)
#时间测试模块
compute_transform(s_perm_mat, P1_gt[:,:,:3], P2_gt[:,:,:3], T1_gt[:,:3,:3], T1_gt[:,:3,3])
infer_time = time.time() - infer_time
match_metrics = matching_accuracy(s_perm_mat, perm_mat, n1_gt)
perform_metrics = compute_metrics(s_perm_mat, P1_gt[:,:,:3], P2_gt[:,:,:3], T1_gt[:,:3,:3], T1_gt[:,:3,3],
viz=viz, usepgm=usepgm, userefine=userefine)
for k in match_metrics:
all_val_metrics_np[k].append(match_metrics[k])
for k in perform_metrics:
all_val_metrics_np[k].append(perform_metrics[k])
all_val_metrics_np['label'].append(Label)
all_val_metrics_np['loss'].append(np.repeat(permevalloss.item(), batch_cur_size))
all_val_metrics_np['infertime'].append(np.repeat(infer_time/batch_cur_size, batch_cur_size))
# 记录详细的对应关系
# all_val_metrics_np['correspond_nomean'].append(s_perm_mat.detach().cpu().numpy())
# all_val_metrics_np['gtcorrespond_nomean'].append(perm_mat.detach().cpu().numpy())
if iter_num % cfg.STATISTIC_STEP == 0 and metric_is_save:
running_speed = cfg.STATISTIC_STEP * batch_cur_size / (time.time() - running_since)
print('Iteration {:<4} {:>4.2f}sample/s'.format(iter_num, running_speed))
running_since = time.time()
all_val_metrics_np = {k: np.concatenate(all_val_metrics_np[k]) for k in all_val_metrics_np}
summary_metrics = summarize_metrics(all_val_metrics_np)
print('Mean-Loss: {:.4f} GT-Acc:{:.4f} Pred-Acc:{:.4f}'.format(summary_metrics['loss'], summary_metrics['acc_gt'], summary_metrics['acc_pred']))
print_metrics(summary_metrics)
if metric_is_save:
np.save(str(Path(cfg.OUTPUT_PATH) / ('eval_log_' + save_filetime + '_metric')),
all_val_metrics_np)
time_elapsed = time.time() - since
print('Evaluation complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
model.train(mode=was_training)
return summary_metrics
def caliters_perm(model, P1_gt_copy, P2_gt_copy, A1_gt, A2_gt, n1_gt, n2_gt, estimate_iters):
# P1_gt_copy_1 = P1_gt_copy.clone()
lap_solver1 = hungarian
s_perm_indexs = []
for estimate_iter in range(estimate_iters):
s_prem_i, Inlier_src_pre, Inlier_ref_pre = model(P1_gt_copy, P2_gt_copy,
A1_gt, A2_gt, n1_gt, n2_gt)
if cfg.PGM.USEINLIERRATE:
s_prem_i = Inlier_src_pre * s_prem_i * Inlier_ref_pre.transpose(2, 1).contiguous()
s_perm_i_mat = lap_solver1(s_prem_i, n1_gt, n2_gt, Inlier_src_pre, Inlier_ref_pre)
# P2_gt_copy_1 = P2_gt_copy.clone()
P2_gt_copy1, s_perm_i_mat_index = calcorrespondpc(s_perm_i_mat, P2_gt_copy)
# Norm2_copy, _ = calcorrespondpc(s_perm_i_mat, Norm2_copy)
s_perm_indexs.append(s_perm_i_mat_index)
if cfg.EXPERIMENT.USERANSAC:
R_pre, T_pre, s_perm_i_mat = RANSACSVDslover(P1_gt_copy[:,:,:3], P2_gt_copy1[:,:,:3], s_perm_i_mat)
else:
R_pre, T_pre = SVDslover(P1_gt_copy[:,:,:3], P2_gt_copy1[:,:,:3], s_perm_i_mat)
P1_gt_copy[:,:,:3] = torch.bmm(P1_gt_copy[:,:,:3], R_pre.transpose(2, 1).contiguous()) + T_pre[:, None, :]
P1_gt_copy[:,:,3:6] = P1_gt_copy[:,:,3:6] @ R_pre.transpose(-1, -2)
# metric, metric_all = calmetrics(s_perm_i_mat, P1_gt_copy_1, P2_gt_copy_1, T1_gt[:, :3, :3], T1_gt[:, :3, 3], viz=viz)
# print(metric_all)
# s_perm_i_mat_sum = torch.sum(s_perm_i_mat, dim=2)
# s_perm = s_perm_indexs[0]
# for s_perm_index in s_perm_indexs[1:]:
# for s_pi in range(s_perm.shape[0]):
# s_perm[s_pi] = s_perm[s_pi, s_perm_index[s_pi]]
# s_pre_prem_mat = np.zeros((A2_gt.shape[0], A1_gt.shape[1], A2_gt.shape[1]))
# for _i in range(A2_gt.shape[0]):
# for row, col in zip(range(A1_gt.shape[1]), s_perm[_i]):
# if s_perm_i_mat_sum[_i, row]:
# s_pre_prem_mat[_i, row, col] = 1
# s_perm_mat = torch.from_numpy(s_pre_prem_mat).to(s_prem_i).contiguous()
return s_perm_i_mat
if __name__ == '__main__':
from utils.dup_stdout_manager import DupStdoutFileManager
from utils.parse_argspc import parse_args
from utils.print_easydict import print_easydict
from utils.visdomshow import VisdomViz
import socket
args = parse_args('Point could registration of graph matching evaluation code.')
import importlib
mod = importlib.import_module(cfg.MODULE)
Net = mod.Net
if cfg.VISDOM.OPEN:
hostname = socket.gethostname()
Visdomins = VisdomViz(env_name=hostname+'PGM_Eval', server=cfg.VISDOM.SERVER, port=cfg.VISDOM.PORT)
Visdomins.viz.close()
else:
Visdomins = None
torch.manual_seed(cfg.RANDOM_SEED)
pc_dataset = get_datasets(partition = 'test',
num_points = cfg.DATASET.POINT_NUM,
unseen = cfg.DATASET.UNSEEN,
noise_type = cfg.DATASET.NOISE_TYPE,
rot_mag = cfg.DATASET.ROT_MAG,
trans_mag = cfg.DATASET.TRANS_MAG,
partial_p_keep = cfg.DATASET.PARTIAL_P_KEEP)
dataloader = get_dataloader(pc_dataset)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model = Net()
model = model.to(device)
model = DataParallel(model, device_ids=cfg.GPUS)
if not Path(cfg.OUTPUT_PATH).exists():
Path(cfg.OUTPUT_PATH).mkdir(parents=True)
now_time = datetime.now().strftime('%Y-%m-%d-%H-%M-%S')
with DupStdoutFileManager(str(Path(cfg.OUTPUT_PATH) / ('eval_log_' + now_time + '.log'))) as _:
print_easydict(cfg)
metrics = eval_model(model, dataloader,
eval_epoch=cfg.EVAL.EPOCH if cfg.EVAL.EPOCH != 0 else None,
metric_is_save=True,
estimate_iters=cfg.EVAL.ITERATION_NUM,
viz=Visdomins,
usepgm=cfg.EXPERIMENT.USEPGM, userefine=cfg.EXPERIMENT.USEREFINE,
save_filetime=now_time)