Vision Mamba 2: More Efficient Visual Representation Learning with State Space Duality
| Dataset | model | date | epoch | top1 | top5 | Checkpoint |
|---|---|---|---|---|---|---|
| ImageNet 1k | DeiT-tiny(baseline) | - | 300 | 72.2 | 91.1 | |
| ImageNet 1k | Vim-tiny (baseline) | - | 300 | 76.1 | 93.0 | |
| ImageNet 1k | Vim2-tiny | 6.11 | 3(training) | 0.106 | - |
We fixed some bugs, the base code is based on Vim and Mamba2, we will update the latest experimental results later.
We are running experiments on imagenet1k and will make the code and weights public later!
If you find Vision mamba2 is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.
@article{vim2,
title={Vision Mamba 2: More Efficient Visual Representation Learning with State Space Duality},
author={Ziwen Chen},
year={2024}
}