File size: 1,966 Bytes
e31dc46 18854f9 07c00e0 18854f9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
---
license: apache-2.0
---
<br>
# Vim Model Card
## Model Details
Vision Mamba (Vim) is a generic backbone trained on the ImageNet-1K dataset for vision tasks.
- **Developed by:** [HUST](https://english.hust.edu.cn/), [Horizon Robotics](https://en.horizon.cc/), [BAAI](https://www.baai.ac.cn/english.html)
- **Model type:** A generic vision backbone based on the bidirectional state space model (SSM) architecture.
- **License:** Non-commercial license
### Model Sources
- **Repository:** https://github.com/hustvl/Vim
- **Paper:** https://arxiv.org/abs/2401.09417
## Uses
The primary use of Vim is research on vision tasks, e.g., classification, segmentation, detection, and instance segmentation, with an SSM-based backbone.
The primary intended users of the model are researchers and hobbyists in computer vision, machine learning, and artificial intelligence.
## How to Get Started with the Model
- You can replace the backbone for vision tasks with the proposed Vim: https://github.com/hustvl/Vim/blob/main/vim/models_mamba.py
- Then you can load this checkpoint and start training.
## Training Details
Vim is pretrained on ImageNet-1K with classification supervision.
The training data is around 1.3M images from [ImageNet-1K dataset](https://www.image-net.org/challenges/LSVRC/2012/).
See more details in this [paper](https://arxiv.org/abs/2401.09417).
## Evaluation
Vim-tiny is evaluated on ImageNet-1K val set, and achieves 76.1% Top-1 Acc. By further finetuning at finer granularity, Vim-tiny achieves 78.3% Top-1 Acc. See more details in this [paper](https://arxiv.org/abs/2401.09417).
## Additional Information
### Citation Information
```
@article{vim,
title={Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model},
author={Lianghui Zhu and Bencheng Liao and Qian Zhang and Xinlong Wang and Wenyu Liu and Xinggang Wang},
journal={arXiv preprint arXiv:2401.09417},
year={2024}
}
```
|