Skip to content

Latest commit

 

History

History
44 lines (30 loc) · 1.92 KB

README.md

File metadata and controls

44 lines (30 loc) · 1.92 KB

Dataset

To download and prepare the MSCOCO 2017 dataset, execute the following commands:

wget http://images.cocodataset.org/zips/train2017.zip -P coco
wget http://images.cocodataset.org/zips/val2017.zip -P coco
wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip -P coco
cd coco
unzip -q train2017.zip
unzip -q val2017.zip
unzip -q annotations_trainval2017.zip
rm train2017.zip
rm val2017.zip
rm annotations_trainval2017.zip

Then in config/FastVim/lsj-100e_coco-instance.py, change data_root path to dataset

Training

chmod +x ./detection/tools/dist_train.sh

FastVimT: ./detection/tools/dist_train.sh "detection/configs/FastVim/vitdet_cascade_mask-rcnn_FastVim_tiny_noclstok_rotate_layernorm_lsj-300e.py" 8 "path_to_imagenet_supervised_ckpt" 

FastVimS: ./detection/tools/dist_train.sh "detection/configs/FastVim/vitdet_cascade_mask-rcnn_FastVim_small_noclstok_rotate_layernorm_lsj-300e.py" 8 "path_to_imagenet_supervised_ckpt" 

FastVimB: ./detection/tools/dist_train.sh "detection/configs/FastVim/vitdet_cascade_mask-rcnn_FastVim_base_noclstok_rotate_layernorm_lsj-300e.py" 8 "path_to_imagenet_supervised_ckpt" 

VimS: ./detection/tools/dist_train.sh "detection/configs/FastVim/vitdet_cascade_mask-rcnn_Vim_small_lsj-300e.py" 8 "path_to_imagenet_supervised_ckpt"

Model weights and configurations

Model APbox
FastVim-T.ckpt 45.1
FastVim-S.ckpt 48.4
FastVim-B.ckpt 50.0
Vim-S.ckpt 47.1