forked from Westlake-AI/MogaNet
-
Notifications
You must be signed in to change notification settings - Fork 0
/
extract_ckpt.py
43 lines (34 loc) · 1.19 KB
/
extract_ckpt.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
"""
Extract parameters in the trained checkpoint.
Example command:
python extract_ckpt.py /path/to/checkpoint /path/to/output.pth.tar
"""
import torch
import argparse
def parse_args():
parser = argparse.ArgumentParser(
description='This script extracts backbone weights from a checkpoint')
parser.add_argument('checkpoint', help='checkpoint file')
parser.add_argument(
'output', type=str, help='destination file name')
parser.add_argument(
'--ema_only',
action='store_true',
help='only ema params as state_dict')
args = parser.parse_args()
return args
def main():
args = parse_args()
assert args.output.endswith(".pth.tar")
ck = torch.load(args.checkpoint, map_location=torch.device('cpu'))
output_dict = dict(state_dict=dict(), author="MogaNet")
keep_keys = ['arch', 'state_dict', 'state_dict_ema', 'metric']
for key in keep_keys:
if ck.get(key, None) is not None:
output_dict[key] = ck[key]
# copy ema params to `state_dict`
if args.ema_only:
output_dict['state_dict'] = output_dict.pop('state_dict_ema')
torch.save(output_dict, args.output)
if __name__ == '__main__':
main()