Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
The official implementation of [CVPR 2025] "5%>100%: Breaking Performance Shackles of Full Fine-Tuning on Visual Recognition Tasks".
| Date | Stars |
|---|---|
| 2026-07-31 | 397 |
| 2026-08-02 | 398 |
| 2026-08-06 | 398 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<div align="center">
## 🔥[CVPR 2025] 5%>100%: Breaking Performance Shackles of Full Fine-Tuning on Visual Recognition Tasks
### [CVPR Homepage](https://cvpr.thecvf.com/virtual/2025/poster/34704) | [PDF](https://openaccess.thecvf.com/content/CVPR2025/papers/Yin_5100_Breaking_Performance_Shackles_of_Full_Fine-Tuning_on_Visual_Recognition_CVPR_2025_paper.pdf)
</div>
- 🔥Mona has been reported by [Extreme Mart](https://mp.weixin.qq.com/s/qL8IzpaFUWKBxUsodshH0w), [PaperWeekly](https://mp.weixin.qq.com/s/OM0hZMX_KcEXvt0DorPhOw), [QbitAI (量子位)](https://mp.weixin.qq.com/s/NsNRmzn_haq_ly63W0WjAQ), and [Synces (机器之心)](https://mp.weixin.qq.com/s/15bO4RD8Iu0TcEfvHGYF0A) !
- 🔥Chinese introduction [【CVPR25】打破全参数微调的性能枷锁!](https://zhuanlan.zhihu.com/p/27268256527)
- 🔥Codes related to DOTA/STAR can be found [here](https://github.com/VisionXLab/STAR-MMRotate/tree/main/configs/h2rbox_v2p/peft), thanks to [@yangxue0827](https://github.com/yangxue0827).
## Table of Contents
- [Introduction](#introduction)
- [Main Results](#main-results)
- [Getting Started](#getting-started)
- [Citation](#citation)
- [Acknowledgement](#acknowledegment)
# Introduction
Pre-training & fine-tuning can enhance the transferring efficiency and performance in visual tasks. Recent deltatuning
methods provide more options for visual classification tasks. Despite their success, existing visual delta-tuning art
fails to exceed the upper limit of full fine-tuning on challenging tasks like instance segmentation and semantic
segmentation. To find a competitive alternative to full fine-tuning, we propose the Multi-cognitive Visual Adapter (Mona)
tuning, a novel adapter-based tuning method.
Mona achieves the strong performance on COCO object detection (`53.4 box AP` and `46.0 mask AP` on `test-dev` with
Swin-Base), and ADE20K semantic segmentation (`51.36 mIoU` on `val` with Swin-Large).
# Main Results
The proposed Mona outperforms full fine-tuning on representative visual tasks, which promotes the upper limit of previous
delta-tuning art. The results demonstrate that the adapter-tuning paradigm can replace full fine-tuning and achieve
better performance in most visual tasks. Full fine-tuning may no longer be the only preferred solution for transfer
learning in the future.
<div align="center">
</div>
Note:
- We report the results with `Cascade Mask R-CNN` (Swin-Base) and `UperNet` (Swin-Large) framework for COCO and ADE20K
respectively.
- The pre-trained weights are IM22K-Supervied pre-trained [Swin-Base](https://drive.google.com/file/d/10-pbubW1-80W0SgWdGP20Q_cm6MlrDpN/view?usp=drive_link
) and [Swin-Large](https://drive.google.com/file/d/142x0CCSItr5p05ea3oylWGaW9MEgnXKU/view?usp=drive_link).
---
Moreover, Mona converges faster than other tested delta-tuning arts.
<div align="center">
</div>
Note:
- We obtain the loss on `VOC` dataset with `RetinaNet` equipped with Swin-Large.
# Getting Started
## Object Detection & Instance Segmentation
### Installation
Please refer to [Swin-Transformer-Object-Detection](Swin-Transformer-Object-Detection/docs/get_started.md) for the
environments and dataset preparation.
### Training Mona
After organizing the dataset, you have to modify the config file according to your environments.
- `data_root`, have to be set as the actual dataset path.
- `load_from`, should be set to your pre-trained weight path.
- `norm_cfg`, have to be set to `SyncBN` if you train the model with multi-gpus.
Please execute the following command in the project path.
#### COCO
```shell
bash Swin-Transformer-Object-Detection/tools/dist_train.sh Swin-Transformer-Object-Detection/mona_configs/swin-b_coco/cascade_mask_swin_base_3x_coco_sample_1_bs_16_mona.py `Your GPUs`
```
#### VOC
```shell
bash Swin-Transformer-Object-Detection/tools/dist_train.sh Swin-Transformer-Object-Detection/mona_configs/swin-l_voc/voc_retinanet_swiExcerpt of 7,916 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:105f3108b901ca7e, desc:fine-tuning, desc:fine tuning