-
Notifications
You must be signed in to change notification settings - Fork 24
/
main_mixste_tpc.py
167 lines (124 loc) · 5.82 KB
/
main_mixste_tpc.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
import os
import sys
import time
import glob
import time
import torch
import random
import socket
import logging
import numpy as np
from tqdm import tqdm
import torch.nn as nn
import torch.utils.data
from common.utils import *
import torch.optim as optim
from common.camera import *
import common.loss as eval_loss
from common.arguments import parse_args
from common.load_data_hm36 import Fusion
from common.h36m_dataset import Human36mDataset
args = parse_args()
os.environ["CUDA_VISIBLE_DEVICES"] = args.gpu
exec('from model.' + args.model + ' import Model')
def train(dataloader, model, optimizer, epoch):
model.train()
loss_all = {'loss': AccumLoss()}
for i, data in enumerate(tqdm(dataloader)):
batch_cam, gt_3D, input_2D, input_2D_GT, action, subject, cam_ind = data
input_2D, input_2D_GT, gt_3D, batch_cam = input_2D.cuda(), input_2D_GT.cuda(), gt_3D.cuda(), batch_cam.cuda()
output_3D, index = model(input_2D)
out_target = gt_3D.clone()
out_target[:, :, args.root_joint] = 0
batch_ind = torch.arange(out_target.shape[0], device=out_target.device).unsqueeze(-1)
out_target = out_target[batch_ind, index]
w_mpjpe = torch.tensor([1, 1, 2.5, 2.5, 1, 2.5, 2.5, 1, 1, 1, 1.5, 1.5, 4, 4, 1.5, 4, 4]).cuda()
loss_w_mpjpe = eval_loss.weighted_mpjpe(output_3D, out_target, w_mpjpe)
loss_temporal = eval_loss.temporal_consistency(output_3D, out_target, w_mpjpe)
loss_mean_velocity = eval_loss.mean_velocity(output_3D, out_target, axis=1)
loss = loss_w_mpjpe + 0.5 * loss_temporal + 2.0 * loss_mean_velocity
optimizer.zero_grad()
loss.backward()
optimizer.step()
N = input_2D.shape[0]
loss_all['loss'].update(loss.detach().cpu().numpy() * N, N)
return loss_all['loss'].avg
def test(actions, dataloader, model):
model.eval()
action_error = define_error_list(actions)
joints_left = [4, 5, 6, 11, 12, 13]
joints_right = [1, 2, 3, 14, 15, 16]
for i, data in enumerate(tqdm(dataloader, dynamic_ncols=True)):
batch_cam, gt_3D, input_2D, input_2D_GT, action, subject, cam_ind = data
input_2D, input_2D_GT, gt_3D, batch_cam = input_2D.cuda(), input_2D_GT.cuda(), gt_3D.cuda(), batch_cam.cuda()
output_3D_non_flip, index_non_flip = model(input_2D[:, 0])
output_3D_flip, index_flip = model(input_2D[:, 1], index_non_flip)
output_3D_flip[:, :, :, 0] *= -1
output_3D_flip[:, :, joints_left + joints_right, :] = output_3D_flip[:, :, joints_right + joints_left, :]
output_3D = (output_3D_non_flip + output_3D_flip) / 2
out_target = gt_3D.clone()
if args.stride == 1:
pad_exist, index_center = torch.where(index_non_flip == args.pad)
index_center = index_center.unsqueeze(1)
batch_ind = torch.arange(output_3D.shape[0], device=output_3D.device).unsqueeze(-1)
output_3D = output_3D[batch_ind, index_center]
out_target = gt_3D[:, args.pad].unsqueeze(1)
else:
batch_ind = torch.arange(output_3D.shape[0], device=output_3D.device).unsqueeze(-1)
out_target = gt_3D[batch_ind, index_non_flip]
output_3D[:, :, args.root_joint] = 0
out_target[:, :, args.root_joint] = 0
action_error = test_calculation(output_3D, out_target, action, action_error, args.dataset, subject)
p1, p2 = print_error(args.dataset, action_error, 1)
return p1, p2
if __name__ == '__main__':
seed = 1126
random.seed(seed)
torch.manual_seed(seed)
np.random.seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
dataset_path = args.root_path + 'data_3d_' + args.dataset + '.npz'
dataset = Human36mDataset(dataset_path, args)
actions = define_actions(args.actions)
if args.train:
train_data = Fusion(args, dataset, args.root_path, train=True)
train_dataloader = torch.utils.data.DataLoader(train_data, batch_size=args.batch_size,
shuffle=True, num_workers=int(args.workers), pin_memory=True)
test_data = Fusion(args, dataset, args.root_path, train=False)
test_dataloader = torch.utils.data.DataLoader(test_data, batch_size=args.batch_size,
shuffle=False, num_workers=int(args.workers), pin_memory=True)
model = Model(args).cuda()
if args.previous_dir != '':
Load_model(args, model)
lr = args.lr
optimizer = optim.AdamW(model.parameters(), lr=lr, weight_decay=0.1)
best_epoch = 0
loss_epochs = []
mpjpes = []
for epoch in range(1, args.nepoch + 1):
if args.train:
loss = train(train_dataloader, model, optimizer, epoch)
loss_epochs.append(loss * 1000)
with torch.no_grad():
p1, p2 = test(actions, test_dataloader, model)
mpjpes.append(p1)
if args.train and p1 < args.previous_best:
best_epoch = epoch
args.previous_name = save_model(args, epoch, p1, model, 'model')
args.previous_best = p1
if args.train:
logging.info('epoch: %d, lr: %.6f, l: %.4f, p1: %.2f, p2: %.2f, %d: %.2f' % (epoch, lr, loss, p1, p2, best_epoch, args.previous_best))
print('%d, lr: %.6f, l: %.4f, p1: %.2f, p2: %.2f, %d: %.2f' % (epoch, lr, loss, p1, p2, best_epoch, args.previous_best))
if epoch % args.lr_decay_epoch == 0:
lr *= args.lr_decay_large
for param_group in optimizer.param_groups:
param_group['lr'] *= args.lr_decay_large
else:
lr *= args.lr_decay
for param_group in optimizer.param_groups:
param_group['lr'] *= args.lr_decay
else:
print('p1: %.2f, p2: %.2f' % (p1, p2))
break