Skip to content

Latest commit

 

History

History
29 lines (20 loc) · 1.38 KB

README.md

File metadata and controls

29 lines (20 loc) · 1.38 KB

Dataset

To download and prepare the ADE20K dataset, execute the following commands:

wget http://data.csail.mit.edu/places/ADEchallenge/ADEChallengeData2016.zip
unzip -q ADEChallengeData2016.zip
rm ADEChallengeData2016.zip

Then in config/base/datasets/ade20k.py, change data_root path to dataset

Training

FastVimT: ./segmentation/tools/dist_train.sh "segmentation/configs/FastVim/uppernet_FastVim_tiny_noclstok_rotate_layernorm_8xb2-160k_ade20k-512x512.py" 8 "path_to_imagenet_supervised_ckpt" 

FastVimS: ./segmentation/tools/dist_train.sh "segmentation/configs/FastVim/uppernet_FastVim_small_noclstok_rotate_layernorm_8xb2-160k_ade20k-512x512.py" 8 "path_to_imagenet_supervised_ckpt" 

FastVimB: ./segmentation/tools/dist_train.sh "segmentation/configs/FastVim/uppernet_FastVim_base_noclstok_rotate_layernorm_8xb2-160k_ade20k-512x512.py" 8 "path_to_imagenet_supervised_ckpt"

Model weights and configurations

Model mIOU
FastVim-T.ckpt 41.8
FastVim-S.ckpt 44.9
FastVim-B.ckpt 47.8