cvrl/train_model.py at main · ChengChen2020/cvrl · 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
import os
import json
import argparse
import numpy as np
from datetime import datetime
from thop import profile, clever_format

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torchvision.datasets import CIFAR10

from tools.cvrlDataset import CIFAR10Pair, train_transform, test_transform
from tools.cvrlTrainer import mocoTrainer, simclrTrainer
from model.model import MoCov1, MoCov2, SimCLRv1, SimCLRv2

parser = argparse.ArgumentParser(description='Contrastive Visual Representation Learning')

# model
parser.add_argument('--model_name', type=str, help='model name', default='mocov1')
parser.add_argument('--epochs', type=int, help='number of epochs', default=200)

# data loader
parser.add_argument('--batch_size', type=int, help='batch size', default=512)

# optimizer
parser.add_argument('--learning_rate', type=float, help='learning rate', default=1e-3)
parser.add_argument('--momentum', type=float, help='optimizer momentum', default=0.9)
parser.add_argument('--weight_decay', type=float, help='weight decay factor', default=1e-6)

# scheduler
parser.add_argument('--step_size', type=float, help='decay lr every step epochs', default=10)
parser.add_argument('--gamma', type=float, help='lr decay factor', default=0.5)
parser.add_argument('--cos', action='store_true', help='use cosine lr schedule')

# backbone
parser.add_argument('--arch', type=str, help='backbone cnn architecture', default='resnet18')

# softmax
parser.add_argument('--temperature', default=0.5, type=float, help='temperature used in softmax')

# moco
parser.add_argument('--moco-dim', default=128, type=int, help='feature dimension')
parser.add_argument('--moco-k', default=4096, type=int, help='queue size; number of negative keys')
parser.add_argument('--moco-m', default=0.99, type=float, help='moco momentum of updating key encoder')

# knn
parser.add_argument('--k', default=200, type=int, help='top k most similar images used to predict the label')

# path
parser.add_argument('--dataset_dir', type=str, default="data/")
parser.add_argument('--log_dir', type=str, help="dir to log", default='train_log/')
parser.add_argument('--results_dir', type=str, help='dir to cache (default: none)', default='')
parser.add_argument('--resume_path', type=str, help='path to latest checkpoint (default: none)', default='')

if __name__ == '__main__':

	args = parser.parse_args()

	if args.results_dir == '':
		args.results_dir = '/cache-' + datetime.now().strftime("%Y-%m-%d-%H-%M-%S-{}".format(args.model_name))

	if not os.path.exists(args.log_dir):
		os.mkdir(args.log_dir)

	train_data = CIFAR10Pair(root=args.dataset_dir, train=True, transform=train_transform, download=True)
	train_iter = DataLoader(train_data, batch_size=args.batch_size, shuffle=True, num_workers=2, pin_memory=True, drop_last=True)

	memory_data = CIFAR10(root=args.dataset_dir, train=True, transform=test_transform, download=True)
	memory_iter = DataLoader(memory_data, batch_size=args.batch_size, shuffle=False, num_workers=2, pin_memory=True)

	test_data = CIFAR10(root=args.dataset_dir, train=False, transform=test_transform, download=True)
	test_iter = DataLoader(test_data, batch_size=args.batch_size, shuffle=False, num_workers=2, pin_memory=True)

	train_log = args.log_dir + args.results_dir

	model_name = args.model_name
	if model_name == 'mocov1':

		args.cos = True
		model = MoCov1(feature_dim=args.moco_dim, K=args.moco_k, m=args.moco_m, T=args.temperature, arch=args.arch, bn_splits=8).cuda()

		optimizer = torch.optim.SGD(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay, momentum=args.momentum)
		trainer = mocoTrainer(train_log, model, train_iter, memory_iter, test_iter, optimizer, args.temperature, args.k, args.learning_rate, args.cos)

	elif model_name == 'mocov2':

		args.cos = True
		model = MoCov2(feature_dim=args.moco_dim, K=args.moco_k, m=args.moco_m, T=args.temperature, arch=args.arch, bn_splits=8).cuda()

		optimizer = torch.optim.SGD(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay, momentum=args.momentum)
		trainer = mocoTrainer(train_log, model, train_iter, memory_iter, test_iter, optimizer, args.temperature, args.k, args.learning_rate, args.cos)

	elif model_name == 'simclrv1':

		model = SimCLRv1(arch=args.arch).cuda()

		# 24.62M 1.31G
		flops, params = profile(model, inputs=(torch.randn(1, 3, 32, 32).cuda(),))
		flops, params = clever_format([flops, params])
		print('# Model Params: {} FLOPs: {}'.format(params, flops))

		optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay)
		trainer = simclrTrainer(train_log, model, train_iter, memory_iter, test_iter, optimizer, args.temperature, args.k)

	elif model_name == 'simclrv2':

		model = SimCLRv2(arch=args.arch).cuda()

		# 26.19M 1.31G
		flops, params = profile(model, inputs=(torch.randn(1, 3, 32, 32).cuda(),))
		flops, params = clever_format([flops, params])
		print('# Model Params: {} FLOPs: {}'.format(params, flops))

		optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate, weight_decay=args.weight_decay)
		trainer = simclrTrainer(train_log, model, train_iter, memory_iter, test_iter, optimizer, args.temperature, args.k)

	else:
		assert(False)

	if not os.path.exists(train_log):
		os.mkdir(train_log)

	config_f = open(os.path.join(train_log, 'config.json'), 'w')
	json.dump(vars(args), config_f)
	config_f.close()

	epoch_start = 1
	trainer.train(args.resume_path, epoch_start, args.epochs)