-
Notifications
You must be signed in to change notification settings - Fork 4
/
engine.py
193 lines (160 loc) · 6.2 KB
/
engine.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
# Copyright (c) V-DETR authors. All Rights Reserved.
import torch
import datetime
import logging
import math
import time
import sys
import numpy as np
from torch.distributed.distributed_c10d import reduce
from utils.ap_calculator import APCalculator
from utils.misc import SmoothedValue
from utils.dist import (
all_gather_dict,
all_reduce_average,
is_primary,
reduce_dict,
barrier,
batch_dict_to_cuda,
)
from utils.box_util import (flip_axis_to_camera_tensor, get_3d_box_batch_tensor)
def compute_learning_rate(args, curr_epoch_normalized):
assert curr_epoch_normalized <= 1.0 and curr_epoch_normalized >= 0.0
if (
curr_epoch_normalized <= (args.warm_lr_epochs / args.max_epoch)
and args.warm_lr_epochs > 0
):
# Linear Warmup
curr_lr = args.warm_lr + curr_epoch_normalized * args.max_epoch * (
(args.base_lr - args.warm_lr) / args.warm_lr_epochs
)
else:
if args.lr_scheduler == 'cosine':
# Cosine Learning Rate Schedule
curr_lr = args.final_lr + 0.5 * (args.base_lr - args.final_lr) * (
1 + math.cos(math.pi * curr_epoch_normalized)
)
else:
step_1, step_2 = args.step_epoch.split('_')
step_1, step_2 = int(step_1), int(step_2)
if curr_epoch_normalized < (step_1 / args.max_epoch):
curr_lr = args.base_lr
elif curr_epoch_normalized < (step_2 / args.max_epoch):
curr_lr = args.base_lr / 10
else:
curr_lr = args.base_lr / 100
return curr_lr
def adjust_learning_rate(args, optimizer, curr_epoch):
curr_lr = compute_learning_rate(args, curr_epoch)
for param_group in optimizer.param_groups:
param_group["lr"] = curr_lr
return curr_lr
def train_one_epoch(
args,
curr_epoch,
model,
optimizer,
criterion,
dataset_config,
dataset_loader,
):
ap_calculator = None
curr_iter = curr_epoch * len(dataset_loader)
max_iters = args.max_epoch * len(dataset_loader)
net_device = next(model.parameters()).device
loss_avg = SmoothedValue(window_size=10)
model.train()
barrier()
for batch_idx, batch_data_label in enumerate(dataset_loader):
curr_time = time.time()
curr_lr = adjust_learning_rate(args, optimizer, curr_iter / max_iters)
batch_data_label = batch_dict_to_cuda(batch_data_label,local_rank=net_device)
# Forward pass
optimizer.zero_grad()
inputs = {
"point_clouds": batch_data_label["point_clouds"],
"point_cloud_dims_min": batch_data_label["point_cloud_dims_min"],
"point_cloud_dims_max": batch_data_label["point_cloud_dims_max"],
}
if args.use_superpoint:
inputs["superpoint_per_point"] = batch_data_label["superpoint_labels"]
outputs = model(inputs)
# Compute loss
loss, loss_dict = criterion(outputs, batch_data_label)
loss_reduced = all_reduce_average(loss)
loss_dict_reduced = reduce_dict(loss_dict)
if not math.isfinite(loss_reduced.item()):
logging.info(f"Loss in not finite. Training will be stopped.")
sys.exit(1)
loss.backward()
if args.clip_gradient > 0:
torch.nn.utils.clip_grad_norm_(model.parameters(), args.clip_gradient)
optimizer.step()
loss_avg.update(loss_reduced.item())
# logging
if is_primary() and curr_iter % args.log_every == 0:
eta_seconds = (max_iters - curr_iter) * (time.time() - curr_time)
eta_str = str(datetime.timedelta(seconds=int(eta_seconds)))
print(
f"Epoch [{curr_epoch}/{args.max_epoch}]; Iter [{curr_iter}/{max_iters}]; Loss {loss_avg.avg:0.2f}; LR {curr_lr:0.2e}; ETA {eta_str}"
)
curr_iter += 1
barrier()
return ap_calculator, curr_iter, curr_lr, loss_avg.avg, loss_dict_reduced
@torch.no_grad()
def evaluate(
args,
curr_epoch,
model,
criterion,
dataset_config,
dataset_loader,
curr_train_iter,
):
# ap calculator is exact for evaluation. This is slower than the ap calculator used during training.
ap_calculator = APCalculator(
dataset_config=dataset_config,
ap_iou_thresh=[0.25, 0.5],
class2type_map=dataset_config.class2type,
no_nms=args.test_no_nms,
args=args
)
curr_iter = 0
net_device = next(model.parameters()).device
num_batches = len(dataset_loader)
loss_avg = SmoothedValue(window_size=10)
model.eval()
barrier()
epoch_str = f"[{curr_epoch}/{args.max_epoch}]" if curr_epoch > 0 else ""
for batch_idx, batch_data_label in enumerate(dataset_loader):
batch_data_label = batch_dict_to_cuda(batch_data_label,local_rank=net_device)
inputs = {
"point_clouds": batch_data_label["point_clouds"],
"point_cloud_dims_min": batch_data_label["point_cloud_dims_min"],
"point_cloud_dims_max": batch_data_label["point_cloud_dims_max"],
}
outputs = model(inputs)
# Compute loss
loss_str = ""
if criterion is not None:
loss, loss_dict = criterion(outputs, batch_data_label)
loss_reduced = all_reduce_average(loss)
loss_dict_reduced = reduce_dict(loss_dict)
loss_avg.update(loss_reduced.item())
loss_str = f"Loss {loss_avg.avg:0.2f};"
else:
loss_dict_reduced = None
if args.cls_loss.split('_')[0] == "focalloss":
outputs["outputs"]["sem_cls_prob"] = outputs["outputs"]["sem_cls_prob"].sigmoid()
outputs["outputs"] = all_gather_dict(outputs["outputs"])
batch_data_label = all_gather_dict(batch_data_label)
if args.axis_align_test:
outputs["outputs"]["box_corners"] = outputs["outputs"]["box_corners_axis_align"]
ap_calculator.step_meter(outputs, batch_data_label)
if is_primary() and curr_iter % args.log_every == 0:
print(
f"Evaluate {epoch_str}; Batch [{curr_iter}/{num_batches}]"
)
curr_iter += 1
barrier()
return ap_calculator, loss_avg.avg, loss_dict_reduced