-
Notifications
You must be signed in to change notification settings - Fork 35
/
Copy pathtrain_new_task_adapters.py
173 lines (140 loc) · 6.81 KB
/
train_new_task_adapters.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
# train_new_task_adapters.py
# created by Sylvestre-Alvise Rebuffi [[email protected]]
# Copyright © The University of Oxford, 2017-2020
# This code is made available under the Apache v2.0 licence, see LICENSE.txt for details
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
import torchvision
import torchvision.transforms as transforms
import models
import os
import time
import argparse
import numpy as np
from torch.autograd import Variable
import imdbfolder_coco as imdbfolder
import config_task
import utils_pytorch
import sgd
parser = argparse.ArgumentParser(description='PyTorch Residual Adapters training')
parser.add_argument('--dataset', default='cifar100', nargs='+', help='Task(s) to be trained')
parser.add_argument('--lr', default=0.1, type=float, help='initial learning rate')
parser.add_argument('--wd', default=1., type=float, help='weight decay for the classification layer')
parser.add_argument('--wd3x3', default=1., type=float, nargs='+', help='weight decay for the 3x3')
parser.add_argument('--wd1x1', default=1., type=float, nargs='+', help='weight decay for the 1x1')
parser.add_argument('--nb_epochs', default=120, type=int, help='nb epochs')
parser.add_argument('--step1', default=80, type=int, help='nb epochs before first lr decrease')
parser.add_argument('--step2', default=100, type=int, help='nb epochs before second lr decrease')
parser.add_argument('--mode', default='parallel_adapters', type=str, help='Task adaptation mode')
parser.add_argument('--proj', default='11', type=str, help='Position of the adaptation module')
parser.add_argument('--dropout', default='00', type=str, help='Position of dropouts')
parser.add_argument('--expdir', default='/scratch/shared/nfs1/srebuffi/exp/dem_learning/tmp/', help='Save folder')
parser.add_argument('--datadir', default='/scratch/local/ramdisk/srebuffi/decathlon/', help='folder containing data folder')
parser.add_argument('--imdbdir', default='/scratch/local/ramdisk/srebuffi/decathlon/annotations/', help='annotation folder')
parser.add_argument('--source', default='/scratch/shared/nfs1/srebuffi/exp/dem_learning/C100_alone/checkpoint/ckptpost11bnresidual11cifar1000.000180607060.t7', type=str, help='Network source')
parser.add_argument('--seed', default=0, type=int, help='seed')
parser.add_argument('--factor', default='1.', type=float, help='Width factor of the network')
args = parser.parse_args()
args.archi ='default'
config_task.mode = args.mode
config_task.proj = args.proj
config_task.factor = args.factor
args.use_cuda = torch.cuda.is_available()
if type(args.dataset) is str:
args.dataset = [args.dataset]
if type(args.wd3x3) is float:
args.wd3x3 = [args.wd3x3]
if type(args.wd1x1) is float:
args.wd1x1 = [args.wd1x1]
if not os.path.isdir(args.expdir):
os.mkdir(args.expdir)
config_task.decay3x3 = np.array(args.wd3x3) * 0.0001
config_task.decay1x1 = np.array(args.wd1x1) * 0.0001
args.wd = args.wd * 0.0001
args.ckpdir = args.expdir + '/checkpoint/'
args.svdir = args.expdir + '/results/'
if not os.path.isdir(args.ckpdir):
os.mkdir(args.ckpdir)
if not os.path.isdir(args.svdir):
os.mkdir(args.svdir)
config_task.isdropout1 = (args.dropout[0] == '1')
config_task.isdropout2 = (args.dropout[1] == '1')
#####################################
# Prepare data loaders
train_loaders, val_loaders, num_classes = imdbfolder.prepare_data_loaders(args.dataset,args.datadir,args.imdbdir,True)
args.num_classes = num_classes
# Load checkpoint and initialize the networks with the weights of a pretrained network
print('==> Resuming from checkpoint..')
checkpoint = torch.load(args.source)
net_old = checkpoint['net']
net = models.resnet26(num_classes)
store_data = []
for name, m in net_old.named_modules():
if isinstance(m, nn.Conv2d) and (m.kernel_size[0]==3):
store_data.append(m.weight.data)
element = 0
for name, m in net.named_modules():
if isinstance(m, nn.Conv2d) and (m.kernel_size[0]==3):
m.weight.data = store_data[element]
element += 1
store_data = []
store_data_bias = []
store_data_rm = []
store_data_rv = []
names = []
for name, m in net_old.named_modules():
if isinstance(m, nn.BatchNorm2d) and 'bns.' in name:
names.append(name)
store_data.append(m.weight.data)
store_data_bias.append(m.bias.data)
store_data_rm.append(m.running_mean)
store_data_rv.append(m.running_var)
# Special case to copy the weight for the BN layers when the target and source networks have not the same number of BNs
import re
condition_bn = 'noproblem'
if len(names) != 51 and args.mode == 'series_adapters':
condition_bn ='bns.....conv'
for id_task in range(len(num_classes)):
element = 0
for name, m in net.named_modules():
if isinstance(m, nn.BatchNorm2d) and 'bns.'+str(id_task) in name and not re.search(condition_bn,name):
m.weight.data = store_data[element].clone()
m.bias.data = store_data_bias[element].clone()
m.running_var = store_data_rv[element].clone()
m.running_mean = store_data_rm[element].clone()
element += 1
#net.linears[0].weight.data = net_old.linears[0].weight.data
#net.linears[0].bias.data = net_old.linears[0].bias.data
del net_old
start_epoch = 0
best_acc = 0 # best test accuracy
results = np.zeros((4,start_epoch+args.nb_epochs,len(args.num_classes)))
all_tasks = range(len(args.dataset))
np.random.seed(1993)
if args.use_cuda:
net.cuda()
cudnn.benchmark = True
# Freeze 3*3 convolution layers
for name, m in net.named_modules():
if isinstance(m, nn.Conv2d) and (m.kernel_size[0]==3):
m.weight.requires_grad = False
args.criterion = nn.CrossEntropyLoss()
optimizer = sgd.SGD(filter(lambda p: p.requires_grad, net.parameters()), lr=args.lr, momentum=0.9, weight_decay=args.wd)
print("Start training")
for epoch in range(start_epoch, start_epoch+args.nb_epochs):
training_tasks = utils_pytorch.adjust_learning_rate_and_learning_taks(optimizer, epoch, args)
st_time = time.time()
# Training and validation
train_acc, train_loss = utils_pytorch.train(epoch, train_loaders, training_tasks, net, args, optimizer)
test_acc, test_loss, best_acc = utils_pytorch.test(epoch,val_loaders, all_tasks, net, best_acc, args, optimizer)
# Record statistics
for i in range(len(training_tasks)):
current_task = training_tasks[i]
results[0:2,epoch,current_task] = [train_loss[i],train_acc[i]]
for i in all_tasks:
results[2:4,epoch,i] = [test_loss[i],test_acc[i]]
np.save(args.svdir+'/results_'+'adapt'+str(args.seed)+args.dropout+args.mode+args.proj+''.join(args.dataset)+'wd3x3_'+str(args.wd3x3)+'_wd1x1_'+str(args.wd1x1)+str(args.wd)+str(args.nb_epochs)+str(args.step1)+str(args.step2),results)
print('Epoch lasted {0}'.format(time.time()-st_time))