Skip to content

Latest commit

 

History

History
executable file
·
37 lines (28 loc) · 1.43 KB

README.md

File metadata and controls

executable file
·
37 lines (28 loc) · 1.43 KB

Transfer VCLR weights format to different frameworks

VCLR to mmaction2

Here, we transfer the vclr weights format to mmaction2 format through three steps:

Step1: download TSN and TSM template weights of mmaction2

Follow the below command to download the template weight files:

cd ~/VCLR/weights
wget https://haofeik-data.s3.amazonaws.com/VCLR/mm_templates/mm_tsn_template.pth
wget https://haofeik-data.s3.amazonaws.com/VCLR/mm_templates/mm_tsm_template.pth

Step2: replace backbone values of the template weights with VCLR weight

For example, vclr weights path is vclr_k400.pth, and mmaction2 template weights path is mm_tsn_template.pth and mm_tsm_template.pth

  • transfer weights for mmaction2 TSN model:

    python vclr2mm.py --mm-weights mm_tsn_template.pth --vclr-weights vclr_k400.pth

  • transfer weights for mmaction2 TSM model:

    python vclr2mm.py --mm-weights mm_tsm_template.pth --vclr-weights vclr_k400.pth --tsm

The transfered weights vclr_mm.pth or vclr_mm_tsm.pth will be stored at the same directory.

VCLR to torchvision

Most applications will use the ResNet-50 through torchvision, you can follow commands below to transfer VCLR pretrained weights to torchvision format.

For example, vclr weights path is vclr_k400.pth,

python vclr2torchvision.py --vclr-weights vclr_k400.pth

The transfered weights vclr_torch.pth will be stored at the same directory.