SAR-Net/pose_manager.py at main · hetolin/SAR-Net · GitHub

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
'''
# -*- coding: utf-8 -*-
# @Project   : code
# @File      : pose_manager.py
# @Software  : PyCharm

# @Author    : hetolin
# @Email     : hetolin@163.com
# @Date      : 2021/11/4 21:25

# @Desciption:
'''

import os

import torch as tc
from torch.utils.data import DataLoader
import dataset as dataloader_nocs

import net_respo.net_sarnet as sarnet

import lib.utils_files as file_utils
import torch
import numpy as np
from torch.optim import lr_scheduler
from lib.scheduler import GradualWarmupScheduler


def setup_env(args):
    os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
    os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_id
    print('Using GPU {}.'.format(args.gpu_id))


def setup_network(args, train_mode=True):
    encoder_decoder = sarnet.EncoderDecoder(args.num_cate)
    encoder_decoder.apply(init_weights)
    net = sarnet.SARNet(encoder_decoder)


    net = net.cuda()
    net.set_mode(train=train_mode)

    net_parameters = net.encoder_decoder.parameters()
    optimizer = tc.optim.Adam(net_parameters, lr=args.lr)

    scheduler_steplr = lr_scheduler.StepLR(optimizer, 4, 0.75)
    scheduler_warmup = GradualWarmupScheduler(optimizer, multiplier=1, total_epoch=args.warmup_epochs, after_scheduler=scheduler_steplr)

    if args.net_recover:
        # resume net, optimizer and scheduler_warmup
        start_epoch = recover_network(args, net, optimizer, scheduler_warmup)

        # resemble these two schedulers
        scheduler_steplr.optimizer = optimizer
        scheduler_warmup.optimizer = optimizer
        scheduler_warmup.after_scheduler = scheduler_steplr
    else:
        save_data = (net, optimizer, scheduler_warmup, 0)
        save_network(save_data, args.net_dump_folder)
        start_epoch = 0

    # parallel
    if args.is_parallel:
        print('=== parallel mode ===')
        print(args.device_id)
        net = torch.nn.DataParallel(net, device_ids=args.device_id)
        optimizer = torch.nn.DataParallel(optimizer, device_ids=args.device_id)
        scheduler_steplr = torch.nn.DataParallel(scheduler_steplr, device_ids=args.device_id)
        scheduler_warmup = torch.nn.DataParallel(scheduler_warmup, device_ids=args.device_id)


    return net, optimizer, scheduler_steplr, scheduler_warmup, start_epoch


def load_dataset(args):
    print('load {} dataset from {}'.format(args.dataset, args.json_file_path))
    dataset = dataloader_nocs.NOCS_DataSet(args)


    def setup_seed(seed):
        torch.manual_seed(seed)
        torch.cuda.manual_seed_all(seed)
        np.random.seed(seed)
        torch.backends.cudnn.deterministic = True #deterministic用来固定内部随机性
        torch.backends.cudnn.benchmark = True #benchmark用在输入尺寸一致，可以加速训练

    # 设置随机数种子
    setup_seed(0)

    dataloader = DataLoader(dataset,
                             batch_size=args.batch_size,
                             shuffle=True,
                             num_workers=args.num_workers,
                             pin_memory=True)

    return dataloader


def save_network(data, folder):
    model, optimizer, scheduler, epoch_num = data
    file_utils.create_folder(folder)

    file_name = os.path.join(folder, '{}.pth'.format(str(epoch_num).zfill(6)))
    print('save network at {}'.format(file_name))
    if hasattr(model, 'module'):
        torch.save({
           'net': model.module.state_dict(),
           'optimizer': optimizer.module.state_dict(),
           'epoch': epoch_num,
           'scheduler': scheduler.module.state_dict()
        }, file_name)
    else:
        torch.save({
           'net': model.state_dict(),
           'optimizer': optimizer.state_dict(),
           'epoch': epoch_num,
           'scheduler': scheduler.state_dict()
        }, file_name)
        # print(optimizer.param_groups[0]['lr'])

def recover_network(args, network, optimizer, scheduler):
    print("Recovering network training from checkpoint: %s (%d epochs)" %
          (args.net_recover_folder, args.net_recover_epoch))

    weight_file = os.path.join(args.net_recover_folder, '{}.pth'.format(str(args.net_recover_epoch).zfill(6)))
    load_parameters(network, optimizer, scheduler, weight_file)
    print("Network recovered correctly")
    start_epoch = args.net_recover_epoch + 1

    return start_epoch

def load_parameters(network, optimizer, scheduler, weight_file):
    if hasattr(network, 'module'):
        network.module.load_state_dict(tc.load(weight_file)['net'])
        optimizer.module.load_state_dict(tc.load(weight_file)['optimizer'])
        scheduler.module.load_state_dict(tc.load(weight_file)['scheduler'])
    else:
        network.load_state_dict(tc.load(weight_file)['net'])
        optimizer.load_state_dict(tc.load(weight_file)['optimizer'])
        scheduler.load_state_dict(tc.load(weight_file)['scheduler'])
        # print(optimizer.param_groups[0]['lr'])

    print("Successfully loaded network parameters from file: {}".format(weight_file))

def init_weights(m):
    classname = m.__class__.__name__
    if classname.find('Conv1d') != -1:
        m.weight.data.normal_(0.0, 0.02)
    elif classname.find('Linear') != -1:
        m.weight.data.normal_(0.0, 0.02)