Skip to content

Latest commit

 

History

History
25 lines (21 loc) · 1.16 KB

README.md

File metadata and controls

25 lines (21 loc) · 1.16 KB

U-Lite

PyTorch code for paper 1M parameters are enough? A lightweight CNN-based model for medical image segmentation, APSIPA 2023.

Update

  • Code is in progress.

Abtract

Deep learning models often have to deal with a trade-off between the need for high accuracy and the desire for low computational cost. In this work, we look for a lightweight U-Net-based model which can remain the same or even achieve better performance for the medical image segmetation, namely U-Lite.

Main highlights:

  • U-Lite ultilizes the criss-cross 7x7 convolutional kernels as the main operator.
  • The model contains only 878K parameters, x35 fewer parameters and x6 faster than UNet.

Citation

@INPROCEEDINGS{10317244,
  author={Dinh, Binh-Duong and Nguyen, Thanh-Thu and Tran, Thi-Thao and Pham, Van-Truong},
  booktitle={2023 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)}, 
  title={1M parameters are enough? A lightweight CNN-based model for medical image segmentation}, 
  year={2023},
  volume={},
  number={},
  pages={1279-1284},
  doi={10.1109/APSIPAASC58517.2023.10317244}}