This repository was archived by the owner on Dec 18, 2024. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
168 lines (146 loc) · 8.44 KB
/
main.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
"""
Author : Pavan Teja Varigonda
"""
import os
import argparse
import pandas as pd
import time
import glob
import json
from dataset_gen import Fashion_Dataset
import torchvision.models as models
from torch.utils.data import DataLoader
from default import Network_
import torch
import torchvision
from torchvision import transforms
from bidict import bidict
def main():
parser = argparse.ArgumentParser(description='Fashion Dataset')
parser.add_argument('--dataset', dest='dataset', default='fashion-dataset', type=str,
help=" select among fashion-dataset / fashion-dataset_small")
parser.add_argument('--loss', dest='loss', default='cse', type=str,
help="Focal / Cross Entropy")
parser.add_argument('--only_finetune', default=False, action='store_true',
help=" if not set trains a model from scratch and finetune")
# parser.add_argument('--lr', default=0.01, action=float,
# help="learning rate to use")
parser.add_argument('--batch_size', type=int, default=2)
parser.add_argument('--model_name', type=str, default='resnet18',
help="resnet18/resnet50")
parser.add_argument('--model_weights', type=str, default=None,
help="Load model for finetune or test")
parser.add_argument('--pretrain_in', default=False, action='store_true',
help="Load Imagenet weights")
# for now just single value
parser.add_argument('--epochs', type=int, default=120,
help="epochs one for pretrain one for finetune")
parser.add_argument('--lr', type=float, default=0.001, help="Learning rate")
parser.add_argument('--finetune_lr', type=float, default=0.001, help="Learning rate")
parser.add_argument('--momentum', type=float, default=0.9)
parser.add_argument('--weight_decay', type=float, default=0)
parser.add_argument('--schedule', nargs="+", type=int, default=[60, 80, 120],)
parser.add_argument('--gpuid', nargs="+", type=int, default=[0],
help="The list of gpuid, ex:--gpuid 3 1. Negative value means cpu-only")
parser.add_argument('--optimizer', type=str, default='SGD', help="SGD|Adam|RMSprop|amsgrad|Adadelta|Adagrad|Adamax ...")
parser.add_argument('--workers', type=int, default=1, help="#Thread for dataloader")
parser.add_argument('--nesterov', default=False, action='store_true', help='nesterov up training phase')
parser.add_argument('--exp_name', dest='exp_name', default='default', type=str,
help="Exp name to be added to the suffix")
parser.add_argument('--add_sampler', type=int, default=0,
help="Sampling for imbalance data")
parser.add_argument('--num_workers', dest='num_workers', default=1, type=int,
help="Num data workers")
parser.add_argument('--debug', default=False, action='store_true',
help='debug')
parser.add_argument('--save_after', default=30, type=int,
help='epochs to save model after')
parser.add_argument('--dir_', dest='dir_', default='/h,', type=str,
help="Exp name to be added to the suffix")
parser.add_argument('--freeze', dest='freeze', default=False, action='store_true',
help="Freeze the pretrain model") # give resnets layer
parser.add_argument('--finetune_freeze', dest='finetune_freeze', default=False, action='store_true',
help="Freeze the finetune model")
parser.add_argument('--gamma', dest='gamma', default=0.2, type=float,
help="Lr drop") # give resnets layer
parser.add_argument('--resize', nargs="+", type=int, default=[224, 224, 2],
help="resize images") # give resnets layer
parser.add_argument('--train_between', dest='train_between', default=False, action='store_true',
help="train between")
parser.add_argument('--freeze_layers', nargs="+", type=str, default=['fc.'],
help="which layers to freeze")
parser.add_argument('--switch_all', type=int, default=0,
help="when to switch to all")
parser.add_argument('--test', dest='test', default='pretrain_test', type=str,
help="when to switch to all")
args = parser.parse_args()
debug = args.debug
print(args.freeze_layers)
datasets_ = {}
# create pretrain gna fine tune datsets
datasets_['train_pt'] = Fashion_Dataset(args.dataset, 'train', dir_=args.dir_, debug=debug, resize=args.resize)
datasets_['test_pt'] = Fashion_Dataset(args.dataset, 'test', dir_=args.dir_, debug=debug, resize=args.resize)
datasets_['train_ft'] = Fashion_Dataset(args.dataset, 'train', dir_=args.dir_, finetune=True, debug=debug, resize=args.resize)
datasets_['test_ft'] = Fashion_Dataset(args.dataset, 'test', dir_=args.dir_, finetune=True, debug=debug, resize=args.resize)
# This dict stores the dataloaders of finetune and pre-train
dataloaders_ = {}
for key, val in datasets_.items():
if 'test' in key:
# Sampler is turned off for the testing phase
dataloaders_[key] = DataLoader(datasets_[key], batch_size=args.batch_size,
num_workers=args.num_workers)
else:
sampler = None
if args.add_sampler:
# Weighted sampler, instance are sampled based on total_count[ class i ] / total_num of examples
dict_samp = datasets_[key].class_count
wts = [ val for keu, val in dict_samp.items()]
wts = [1 / wt if wt else 0 for wt in wts ]
wts = torch.FloatTensor(wts)
# weights for the samples
sample_wts = [wts[t] for t in datasets_[key].data['class']]
sampler = torch.utils.data.WeightedRandomSampler(sample_wts, len(sample_wts))
dataloaders_[key] = DataLoader(datasets_[key], batch_size=args.batch_size,
sampler=sampler, num_workers=args.num_workers)
else:
# Suffle only no sampler
dataloaders_[key] = DataLoader(datasets_[key], batch_size=args.batch_size,
sampler=sampler, num_workers=args.num_workers, shuffle=True)
# all parameters passed to the network
config = {'loaders' : {'pretrain' : [dataloaders_['train_pt'], dataloaders_['test_pt']] ,
'finetune' : [dataloaders_['train_ft'], dataloaders_['test_ft']]},
'gpuid': args.gpuid, 'lr': args.lr, 'momentum': args.momentum, 'weight_decay': args.weight_decay,'schedule': args.schedule,
'optimizer':args.optimizer, 'exp_name' : args.exp_name, 'nesterov':args.nesterov, 'model_name': args.model_name, 'pretrain_in':args.pretrain_in,
'model_weights':args.model_weights, 'loss': args.loss, 'save_after':args.save_after, 'freeze': args.freeze, 'gamma' :args.gamma, 'debug':args.debug, 'finetune_lr':args.finetune_lr,
'finetune_freeze':args.finetune_freeze, 'train_between' : args.train_between, 'freeze_layers': args.freeze_layers, 'switch_all':args.switch_all }
if debug:
for key, val in dataloaders_.items():
check_data(val, args.batch_size, key)
net = Network_(config)
if not args.model_weights and not args.only_finetune:
# pre train
net.train_(args.epochs)
else:
print('------------------SKIPPING THE PRETRAN---------------')
if args.test == 'finetune_test':
net.str_ = 'finetune'
# import pdb; pdb.set_trace()
net.switch_finetune()
net.load_model()
acc, acc_5, acc_cl_1, acc_cl_5, losses = net.validation(net.test_loader, 0)
elif args.test == 'pretrain_test':
net.str_ = 'pretrain'
net.load_model()
acc, acc_5, acc_cl_1, acc_cl_5, losses = net.validation(net.test_loader, 0)
else:
net.load_model()
net.train_(args.epochs, finetune=True)
dict_json = {'acc_1': acc.avg, 'acc_5': acc_5.avg, 'acc_cl_1' : acc_cl_1, 'acc_cl_5':acc_cl_5 }
print(dict_json)
import json
str_p = args.model_weights[:args.model_weights.rfind('/')]
str_name = str_p[str_p.rfind('/')+1:]
with open(f'{str_name}_result.json', 'w') as fp:
json.dump(dict_json, fp)
if __name__ == '__main__':
main()