HierAttn
1.0.0
深度監督皮膚病變診斷階段和分支注意力
Wei Dai,Rui Liu,Tianyi Wu,Min Wang,Jianqin Yin,Jun Liu
在IEEE JBHI中接受,2023年。 [Arxiv] [Paper]

請參閱install.md以獲取安裝說明。
請參閱DATA.MD以獲取數據集準備。
我們使用轉移學習來部分初始化Hierattn和Sota模型的可調權重。請參閱premodel.md,以獲取驗證的模型下載。
| ihisic20000 val | #參數/m | TOP-1精度/%↑ |
|---|---|---|
| Mobilenetv2 | 2.23 | 93.45 |
| MobileVit_s | 4.94 | 94.72 |
| mobilenetv3_large | 4.21 | 94.77 |
| ShuffLenetV2_1× | 2.28 | 95.23 |
| mnasnet1.0 | 3.11 | 95.45 |
| 效率網絡_b0 | 4.02 | 95.48 |
| hierattn_xs(我們的) | 1.08 | 96.15 |
| hierattn_s(我們的) | 2.14 | 96.70 |
| IHPAD3000 VAL | #參數/m | TOP-1精度/%↑ |
|---|---|---|
| Mobilenetv2 | 2.23 | 87.44 |
| ShuffLenetV2_1× | 2.28 | 87.89 |
| MobileVit_s | 4.94 | 88.22 |
| mobilenetv3_large | 4.21 | 88.78 |
| 效率網絡_b0 | 4.02 | 90.22 |
| mnasnet1.0 | 3.11 | 90.33 |
| hierattn_xs(我們的) | 1.08 | 90.11 |
| hierattn_s(我們的) | 2.13 | 91.22 |
如果您發現它在研究中很有用,請考慮以下參考文章:
@ARTICLE{10230242,
author={Dai, Wei and Liu, Rui and Wu, Tianyi and Wang, Min and Yin, Jianqin and Liu, Jun},
journal={IEEE Journal of Biomedical and Health Informatics},
title={Deeply Supervised Skin Lesions Diagnosis With Stage and Branch Attention},
year={2024},
volume={28},
number={2},
pages={719-729},
keywords={Skin;Lesions;Feature extraction;Convolution;Transformers;Training;Computational modeling;Attention;deep supervision;disease classification;skin lesion;vision transformer},
doi={10.1109/JBHI.2023.3308697}}
非常感謝ML-CVNets的作者的出色框架!