-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathvalidate_ots_sslbackbone.py
251 lines (199 loc) · 9.94 KB
/
validate_ots_sslbackbone.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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import argparse
import os
import pandas as pd
import time
import logging
import torch
import torch.nn as nn
import torch.nn.parallel
from collections import OrderedDict
from contextlib import suppress
from torchvision import transforms
from torchvision.datasets import ImageNet
from torchvision.transforms.functional import InterpolationMode
import timm
from timm.utils import accuracy, AverageMeter, setup_default_logging
import SSLBackbone_setup.token_reduc as token_reduc
import SSLBackbone_setup.models_vit_mae as mae_models
torch.backends.cudnn.benchmark = True
_logger = logging.getLogger('validate')
parser = argparse.ArgumentParser(description='PyTorch ImageNet Validation')
parser.add_argument('--data', type=str, help='dataset path')
parser.add_argument('--split', metavar='NAME', default='val',
help='dataset split (default: val)')
parser.add_argument('--model-type', help='Model stype [AugReg, SWAG, MAE]', default="augreg", choices=["augreg", "swag", "mae"])
parser.add_argument('--model-name', '-m', metavar='NAME', help='backbone model', default="vit_tiny_patch16_224")
parser.add_argument('--input-size', type=int, default=224)
parser.add_argument('--r', type=int, nargs='+', default=[0])
parser.add_argument('--r-sched', type=str, default="constant", choices=["constant", "decrease"])
parser.add_argument('--merge_fn', type=str, default="tome", choices=["tome", "atc"])
parser.add_argument('--linkage', type=str, default="average", choices=["average","single","complete"])
parser.add_argument('-j', '--num_workers', default=4, type=int, metavar='N',
help='number of data loading workers (default: 2)')
parser.add_argument('-b', '--batch-size', default=256, type=int,
metavar='N', help='mini-batch size (default: 256)')
parser.add_argument('--log-freq', default=10, type=int,
metavar='N', help='batch logging frequency (default: 10)')
parser.add_argument('--pin-mem', action='store_true', default=False,
help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--use_amp', action='store_true', default=False,
help='Use AMP mixed precision. Defaults to Apex, fallback to native Torch AMP.')
parser.add_argument('--device', default='cuda', help='device to use for training / testing')
parser.add_argument('--results-file', default='', type=str, metavar='FILENAME',
help='Output csv file for validation results (summary)')
parser.add_argument('--results-dir', default='./tome_notrain_results', type=str, metavar='DIRBANE',
help='Output directory for validation results (summary)')
# AugReg models:
## vit_tiny_patch16_224 -- r [0, 16] -- im 224
## vit_small_patch16_224 -- r [0, 16] -- im 224
## vit_base_patch16_224 -- r[0, 16] -- im 224
## vit_large_patch16_224 -- r[0, 8] -- im 224
## vit_large_patch16_384 -- r [0, 5 ,10, 15, 20, 23] -- im 384
def timm_prep(model_name = "vit_base_patch16_224", input_size = 224):
# Load a pretrained model
model = timm.create_model(model_name, pretrained=True)
input_size = model.default_cfg["input_size"][1]
transform = transforms.Compose([
transforms.Resize(int((256 / 224) * input_size), interpolation=InterpolationMode.BICUBIC),
transforms.CenterCrop(input_size),
transforms.ToTensor(),
transforms.Normalize(model.default_cfg["mean"], model.default_cfg["std"]),
])
return model, transform, token_reduc.patch.timm, (model.default_cfg["mean"], model.default_cfg["std"])
# SWAG models:
## vit_b16_in1k -- r [0, 5, 10, 15, 20, 25, 30, 35, 40, 45] -- im 384
## vit_l16_in1k -- r [0, 5, 10, 15, 20, 25, 30, 35, 40] -- im 512
## vit_h14_in1k -- r [0, 5, 10, 15, 20, 25, 30, 35, 40] -- im 518
def swag_prep(model_name = "vit_b16_in1k", input_size = 384):
# Load a pretrained model
model = torch.hub.load("facebookresearch/swag", model=model_name)
transform = transforms.Compose([
transforms.Resize(input_size, interpolation=InterpolationMode.BICUBIC),
transforms.CenterCrop(input_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
return model, transform, token_reduc.patch.swag, ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
# MAE models:
## vit_base_patch16 -- r [0, 16] -- im 224
## vit_large_patch16 -- r [0, 8] -- im 224
## vit_huge_patch14 -- r [0, 7] -- im 224
def mae_prep(model_name = "vit_base_patch16", input_size = 224):
model = mae_models.__dict__[model_name](
num_classes=1000,
drop_path_rate=0.0,
global_pool=True,
)
model = mae_models.prepare_model(model, model_name, "./pretrained_models")
if input_size <= 224:
crop_pct = 224 / 256
else:
crop_pct = 1.0
size = int(input_size / crop_pct)
transform = transforms.Compose([
transforms.Resize(size, interpolation=InterpolationMode.BICUBIC),
transforms.CenterCrop(input_size),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
return model, transform, token_reduc.patch.mae, ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
def main():
setup_default_logging()
args = parser.parse_args()
amp_autocast = suppress # do nothing
if args.use_amp:
amp_autocast = torch.cuda.amp.autocast
_logger.info('Validating in mixed precision with native PyTorch AMP.')
device = torch.device(args.device)
if args.model_type == "augreg":
model, transform, patch_fn, norm = timm_prep(args.model_name, args.input_size)
elif args.model_type == "swag":
model, transform, patch_fn, norm = swag_prep(args.model_name, args.input_size)
elif args.model_type == "mae":
model, transform, patch_fn, norm = mae_prep(args.model_name, args.input_size)
patch_fn(model, args.merge_fn, args.linkage)
model.r = 0
dataset_val = ImageNet(root=args.data, split=args.split, transform=transform)
sampler_val = torch.utils.data.SequentialSampler(dataset_val)
data_loader_val = torch.utils.data.DataLoader(
dataset_val, sampler=sampler_val,
batch_size=int(args.batch_size),
num_workers=args.num_workers,
pin_memory=args.pin_mem,
drop_last=False
)
param_count = sum([m.numel() for m in model.parameters()])
_logger.info(f'Model {args.model_type} {args.model_name} created, param count: {param_count}')
model.to(device)
model.eval()
criterion = nn.CrossEntropyLoss().to(device)
args.results_dir = os.path.join(args.results_dir, args.r_sched)
os.makedirs(args.results_dir, exist_ok=True)
if args.results_file == "":
args.results_file = f"{args.model_type}_{args.model_name}_{args.input_size}_{args.merge_fn}"
if args.merge_fn != "tome":
args.results_file += f"_{args.linkage}"
args.results_file += ".csv"
results_dict = {"Model":[],
"Img_Size":[],
"r":[],
"Top-1":[],
"Top-5":[],
"Param":[]}
for r in args.r:
if r != 0:
if args.r_sched == "constant":
model.r = r
elif args.r_sched == "decrease":
model.r = (r, -1)
_logger.info(f"{model.r}, {r}")
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
model.eval()
with torch.no_grad():
end = time.time()
for batch_idx, (input, target) in enumerate(data_loader_val):
target = target.to(device, non_blocking=True)
input = input.to(device, non_blocking=True)
# compute output
with amp_autocast():
output = model(input)
loss = criterion(output, target)
losses.update(loss.item(), input.size(0))
# measure accuracy and record loss
acc1, acc5 = accuracy(output, target, topk=(1, 5))
top1.update(acc1.item(), input.size(0))
top5.update(acc5.item(), input.size(0))
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if batch_idx % args.log_freq == 0:
_logger.info(
'Test: [{0:>4d}/{1}] '
'Time: {batch_time.val:.3f}s ({batch_time.avg:.3f}s, {rate_avg:>7.2f}/s) '
'Loss: {loss.val:>7.4f} ({loss.avg:>6.4f}) '
'Acc@1: {top1.val:>7.3f} ({top1.avg:>7.3f}) '
'Acc@5: {top5.val:>7.3f} ({top5.avg:>7.3f})'.format(
batch_idx, len(data_loader_val), batch_time=batch_time,
rate_avg=input.size(0) / batch_time.avg,
loss=losses, top1=top1, top5=top5))
top1a, top5a = top1.avg, top5.avg
results = OrderedDict(
model = args.model_name, img_size = args.input_size,
top1=round(top1a, 4), top1_err=round(100 - top1a, 4),
top5=round(top5a, 4), top5_err=round(100 - top5a, 4),
param_count=round(param_count / 1e6, 2))
_logger.info(' * Acc@1 {:.3f} ({:.3f}) Acc@5 {:.3f} ({:.3f})'.format(
results['top1'], results['top1_err'], results['top5'], results['top5_err']))
results_dict["Model"].append(args.model_name)
results_dict["Img_Size"].append(args.input_size)
results_dict["r"].append(r)
results_dict["Top-1"].append(top1a)
results_dict["Top-5"].append(top5a)
results_dict["Param"].append(round(param_count / 1e6, 2))
df_results = pd.DataFrame(results_dict)
df_results.to_csv(os.path.join(args.results_dir, args.results_file), sep=";", index=False)
if __name__ == '__main__':
main()