-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain_dp_simsiam.py
195 lines (158 loc) · 6.39 KB
/
train_dp_simsiam.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
# ------------------------------------------------
import cv2 # first import cv2, then torch
cv2.setNumThreads(0)
# https://github.com/pytorch/pytorch/issues/1838
# ------------------------------------------------
import os
import argparse
import torch
import numpy as np
import torch.nn as nn
import torch.optim as optim
import copy
from torch.optim import lr_scheduler
from torch.utils.data import DataLoader
from itertools import chain
from loss_moco import global_features, momentum_update
from scannet.scannet_pretrain import ScannetDepthPointDataset
from model.cnn2d.conv2d import UNet
from model.pointnet2.backbone_module import Pointnet2Backbone
from meters import ProgressMeter
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
class SimSiam(nn.Module):
def __init__(self,
encoder_d=UNet(),
encoder_p=Pointnet2Backbone(),
dim=1024,
pred_dim=256,
momentum_encoder=False,
momentum=0.999):
super(SimSiam, self).__init__()
self.encoder_d = encoder_d
self.encoder_p = encoder_p
self.projector_d = nn.Sequential(
nn.Linear(pred_dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, dim, bias=False),
nn.BatchNorm1d(dim, affine=False)
)
self.projector_p = nn.Sequential(
nn.Linear(pred_dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, dim, bias=False),
nn.BatchNorm1d(dim, affine=False)
)
self.predictor_d = nn.Sequential(
nn.Linear(dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, dim)
)
self.predictor_p = nn.Sequential(
nn.Linear(dim, pred_dim, bias=False),
nn.BatchNorm1d(pred_dim),
nn.ReLU(True),
nn.Linear(pred_dim, dim)
)
self.momentum_encoder = momentum_encoder
if momentum_encoder:
self.m_encoder_d = copy.deepcopy(encoder_d)
self.m_encoder_p = copy.deepcopy(encoder_p)
self.m_projector_d = copy.deepcopy(self.projector_d)
self.m_projector_p = copy.deepcopy(self.projector_p)
self.m_encoder_d.requires_grad_(False)
self.m_encoder_p.requires_grad_(False)
self.m_projector_d.requires_grad_(False)
self.m_projector_p.requires_grad_(False)
self.m = momentum
self.criterion = nn.CosineSimilarity(1)
def sim_siam_forward(self, depth, points):
zd = self.encoder_d(depth)
zd = global_features(zd)
zd = self.projector_d(zd)
zp = self.encoder_p(points)
zp = zp["fp2_features"]
zp = global_features(zp)
zp = self.projector_p(zp)
pd = self.predictor_d(zd)
pp = self.predictor_p(zp)
# only cross format
loss = self.criterion(pd, zp.detach()).mean() + self.criterion(pp, zd.detach()).mean()
loss = loss * 0.5
return loss
def byol_forward(self, depth, points):
zd = self.encoder_d(depth)
zd = global_features(zd)
zd = self.projector_d(zd)
zp = self.encoder_p(points)
zp = zp["fp2_features"]
zp = global_features(zp)
zp = self.projector_p(zp)
pd = self.predictor_d(zd)
pp = self.predictor_p(zp)
with torch.no_grad():
mzd = self.m_encoder_d(depth)
mzd = global_features(mzd)
mzd = self.m_projector_d(mzd)
mzp = self.m_encoder_p(points)
mzp = mzp["fp2_features"]
mzp = global_features(mzp)
mzp = self.m_projector_p(mzp)
loss = self.criterion(pd, mzp).mean() + self.criterion(pp, mzd).mean()
loss *= 0.5
momentum_update(self.encoder_d, self.m_encoder_d, self.m)
momentum_update(self.encoder_p, self.m_encoder_p, self.m)
momentum_update(self.projector_d, self.m_projector_d, self.m)
momentum_update(self.projector_p, self.m_projector_p, self.m)
return loss
def forward(self, depth, points):
if self.momentum_encoder:
return self.byol_forward(depth, points)
else:
return self.sim_siam_forward(depth, points)
def main():
# hyperparamterts
BATCH_SIZE = 6
EPOCH = 100
DEVICE = "cuda:0"
LR = 0.05 / 256 * BATCH_SIZE
MOMENTUM_ENCODER = True
BASE_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "log")
SAVE = os.path.join(BASE_DIR, "DP_SimSiam")
if not os.path.exists(SAVE):
os.mkdir(SAVE)
ds = ScannetDepthPointDataset("train", diff_crop=True, augment=True, num_match=10) # smalle num_match for speed
dl = DataLoader(ds, batch_size=BATCH_SIZE, shuffle=True, num_workers=6)
encoder_d = UNet()
encoder_p = Pointnet2Backbone()
net = SimSiam(encoder_d, encoder_p, dim=1024, pred_dim=256, momentum_encoder=MOMENTUM_ENCODER).to(DEVICE)
optimizer = optim.SGD(net.parameters(), lr=LR, weight_decay=1e-4,
momentum=0.9)
scheduler = lr_scheduler.CosineAnnealingLR(optimizer, EPOCH)
for epoch in range(EPOCH):
net.train()
progress = ProgressMeter(len(dl), prefix="Epoch: [{}]".format(epoch))
for i, data in enumerate(dl):
points = data["pcd"].to(DEVICE)
depth = data["depthmap"].to(DEVICE)
loss = net(depth=depth, points=points)
progress.update(i, {"loss": loss.item()})
optimizer.zero_grad()
loss.backward()
optimizer.step()
if i % 20 == 0:
progress.display(i)
scheduler.step()
torch.save({'epoch': epoch,
'model_state_dict': net.state_dict()},
os.path.join(SAVE,'checkpoint_{:03d}.pth'.format(epoch)))
if __name__ == "__main__":
main()