search

hustvl / ViTMatte

[Information Fusion (Vol.103, Mar. '24)] Boosting Image Matting with Pretrained Plain Vision Transformers
MIT License
339 stars 33 forks source link

finetuning from DINO weights #5

Closed kshitijagrwl closed 1 year ago

kshitijagrwl commented 1 year ago

In your train.md doc, you mention about pretrained folder with a preprocess.py script to convert DINO weights for loading into your architecture. 

cd ViTMatte/pretrained
python preprocess.py

Can you please release this script?

JingfengYao commented 1 year ago

Hi, we have updated the script. You could run 

cd ./pretrained
python preprocess.py

to get dino_vit_s_fna.pth and mae_vit_b_fna.pth for the training.

kshitijagrwl commented 1 year ago

thanks!