mirror of https://github.com/open-mmlab/mmsegmentation.git synced 2025-06-03 22:03:48 +08:00

History

谢昕辰 ef4b30038f [Feature] add DPT head (#605 )

* add DPT head

* [fix] fix init error

* use mmcv function

* delete code

* remove transpose clas

* support NLC output shape

* Delete post_process_layer.py

* add unittest and docstring

* rename variables

* fix project error and add unittest

* match dpt weights

* add configs

* fix vit pos_embed bug and dpt feature fusion bug

* match vit output

* fix gelu

* minor change

* update unitest

* fix configs error

* inference test

* remove auxilary

* use local pretrain

* update training results

* update yml

* update fps and memory test

* update doc

* update readme

* add yml

* update doc

* remove with_cp

* update config

* update docstring

* remove dpt-l

* add init_cfg and modify readme.md

* Update dpt_vit-b16.py

* zh-n README

* use constructor instead of build function

* prevent tensor being modified by ConvModule

* fix unittest

Co-authored-by: Junjun2016 <hejunjun@sjtu.edu.cn>

2021-08-30 16:53:05 +08:00

dpt_vit-b16_512x512_160k_ade20k.py

[Feature] add DPT head (#605 )

2021-08-30 16:53:05 +08:00

dpt.yml

[Feature] add DPT head (#605 )

2021-08-30 16:53:05 +08:00

README.md

[Feature] add DPT head (#605 )

2021-08-30 16:53:05 +08:00

README.md

Vision Transformer for Dense Prediction

Introduction

@article{dosoViTskiy2020,
  title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author={DosoViTskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and  Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil},
  journal={arXiv preprint arXiv:2010.11929},
  year={2020}
}

@article{Ranftl2021,
  author    = {Ren\'{e} Ranftl and Alexey Bochkovskiy and Vladlen Koltun},
  title     = {Vision Transformers for Dense Prediction},
  journal   = {ArXiv preprint},
  year      = {2021},
}

Usage

To use other repositories' pre-trained models, it is necessary to convert keys.

We provide a script vit2mmseg.py in the tools directory to convert the key of models from timm to MMSegmentation style.

python tools/model_converters/vit2mmseg.py ${PRETRAIN_PATH} ${STORE_PATH}

E.g.

python tools/model_converters/vit2mmseg.py https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_p16_224-80ecf9dd.pth pretrain/jx_vit_base_p16_224-80ecf9dd.pth

This script convert model from PRETRAIN_PATH and store the converted model in STORE_PATH.

Results and models

ADE20K

Method	Backbone	Crop Size	Lr schd	Mem (GB)	Inf time (fps)	mIoU	mIoU(ms+flip)	config	download
DPT	ViT-B	512x512	160000	8.09	10.41	46.97	48.34	config	model \| log