mmselfsup/tools/analysis_tools/count_parameters.py

41 lines
1.3 KiB
Python

# Copyright (c) OpenMMLab. All rights reserved.
import argparse
from mmcv import Config
from mmselfsup.models import build_algorithm
def parse_args():
parser = argparse.ArgumentParser(description='Count model parameters')
parser.add_argument('config', help='train config file path')
args = parser.parse_args()
return args
def main():
args = parse_args()
cfg = Config.fromfile(args.config)
model = build_algorithm(cfg.model)
num_params = sum(p.numel() for p in model.parameters()) / 1000000.
num_grad_params = sum(
p.numel() for p in model.parameters() if p.requires_grad) / 1000000.
num_backbone_params = sum(p.numel()
for p in model.backbone.parameters()) / 1000000.
num_backbone_grad_params = sum(p.numel()
for p in model.backbone.parameters()
if p.requires_grad) / 1000000.
print(f'Number of backbone parameters: {num_backbone_params:.5g} M')
print(f'Number of backbone parameters requiring grad: '
f'{num_backbone_grad_params:.5g} M')
print(f'Number of total parameters: {num_params:.5g} M')
print(f'Number of total parameters requiring grad: '
f'{num_grad_params:.5g} M')
if __name__ == '__main__':
main()