Segmentation.md

Plain VisionLLaMA

VisionLLaMA-B with 800 epochs using MAE.

cd mmseg 
bash ./tools/dist_train.sh  configs/mae/lama-base_upernet_8xb2-800-amp-160k_ade20k-512x512.py    8 --amp &> lama-base_upernet_8xb2-800-amp-160k_ade20k-512x512.log

VisionLLaMA-B with 1600 epochs using MAE.

cd mmseg 
bash ./tools/dist_train.sh  configs/mae/lama-base_upernet_8xb2-1600-amp-160k_ade20k-512x512.py    8 --amp &> lama-base_upernet_8xb2-1600-amp-160k_ade20k-512x512.log

Pyramid VisionLLaMA

Pyramid VisionLLaMA-B

cd mmseg 
bash ./tools/dist_train.sh  configs/twins/twins_svt_lama_as-b_uperhead_8xb2-160k_ade20k-512x512.py 8 --amp &> twins_svt_lama_as-b_uperhead_8xb2-160k_ade20k-512x512.log

Pyramid VisionLLaMA-L

cd mmseg 
bash ./tools/dist_train.sh  configs/twins/twins_svt_lama_as-l_uperhead_8xb2-160k_ade20k-512x512.py 8 --amp &> twins_svt_lama_as-l_uperhead_8xb2-160k_ade20k-512x512.log

Acknowledgement

Our code is based on mmsegmentation and Twins. Thanks for their great work. Specifically, we start from this commit id.