search

open-mmlab / mmaction2

OpenMMLab's Next Generation Video Understanding Toolbox and Benchmark
https://mmaction2.readthedocs.io
Apache License 2.0
4.23k stars 1.24k forks source link

Freeze the previous layer #1607

Closed WEIZHIHONG720 closed 2 years ago

WEIZHIHONG720 commented 2 years ago

Hi, When using fine-tuning, how do I freeze the previous layers and only fine-tune the full connection? Thank you ! I use the ipcsn_ig65m_pretrained_bnfrozen_r152_32x2x1_58e_kinetics400_rgb.py :

model = dict( backbone=dict( norm_eval=True, bn_frozen=True, bottleneck_mode='ip', pretrained= # noqa: E251 'https://download.openmmlab.com/mmaction/recognition/csn/ipcsn_from_scratch_r152_ig65m_20210617-c4b99d38.pth' # noqa: E501 ), cls_head=dict( num_classes=400, dropout_ratio=0.6, ) )

And how to freeze the previous laye except for the fully connected layer. Thank you !

kennymckormick commented 2 years ago

Hi, weizh, you can set frozen_stages=4 in the backbone config to do so.