|
--- |
|
license: mit |
|
--- |
|
|
|
|
|
|
|
<br> |
|
|
|
|
|
|
|
# GroupMamba-Small Model Card |
|
|
|
|
|
|
|
## Model Details |
|
|
|
|
|
|
|
GroupMamba-Small is a generic backbone with 34M parameters trained on the ImageNet-1K dataset for vision tasks. |
|
|
|
|
|
|
|
- **Model type:** Parameter-Efficient and Accurate Vision Backbone Based on Group Visual State Space Model |
|
- **License:** Non-commercial license |
|
|
|
|
|
|
|
|
|
|
|
### Model Sources |
|
|
|
|
|
|
|
- **Repository:** https://github.com/amshaker/GroupMamba |
|
- **Paper:** https://arxiv.org/abs/X.X |
|
|
|
|
|
|
|
## Uses |
|
|
|
|
|
|
|
The primary use of GroupMamba 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 GroupMamba: https://github.com/Amshaker/GroupMamba/blob/main/classification/models/groupmamba.py |
|
- Then you can load this checkpoint and start fine-tuning. |
|
|
|
|
|
|
|
## Training Details |
|
|
|
|
|
|
|
GroupMamba 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/X.X. |
|
|
|
|
|
|
|
## Evaluation |
|
|
|
|
|
|
|
GroupMamba-Small is evaluated on ImageNet-1K val set, and achieves 83.9% Top-1 Acc with only 34M parameters. See more details in this [paper](https://arxiv.org/abs/X.X). |
|
|
|
|
|
|
|
## Additional Information |
|
|
|
|
|
|
|
### Citation Information |
|
|
|
|
|
|
|
``` |
|
@article{GroupMamba, |
|
title={GroupMamba: Parameter-Efficient and Accurate Group Visual State Space Model}, |
|
author={Abdelrahman Shaker and Syed Talal Wasim and Salman Khan and Gall Jürgen and Fahad Khan}, |
|
journal={arXiv preprint arXiv:X.X}, |
|
year={2024} |
|
} |
|
``` |
|
|
|
|
|
|
|
|