Hugging Face's logo

yxlu0
/
SAK

Model card Files Community
SAK
File size: 2,531 Bytes 
4c33a99
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
60
61
62
63
64
65
66
67
68
69
70
 
---
license: mit
datasets:
- ILSVRC/imagenet-1k
---

# SAK

<!-- Provide a quick summary of what the model is/does. -->

These are checkpoints for our ICLR2025 paper: **Swiss Army Knife: Synergizing Biases in Knowledge from Vision Foundation Models for Multi-Task Learning**.

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->



- **Developed by:** Yuxiang Lu, Shengcao Cao, Yu-Xiong Wang
- **License:** mit

### Model Sources

<!-- Provide the basic links for the model. -->

- **Repository:** https://github.com/innovator-zero/SAK
- **Paper [OpenReview]:** https://openreview.net/forum?id=eePww5u7J3
- **Paper [arXiv]:** https://arxiv.org/abs/2410.14633
- **Project Page:** https://innovator-zero.github.io/SAK/

## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

Currently we directly provide checkpoints of pre-trained models in this repository. For detailed information on usage, please refer to our [github repository](https://github.com/innovator-zero/SAK).

Following are the checkpoint lists:

**Stage 1**
| Teachers                | Student backbone | Checkpoint |
| ----------------------- | ---------------- | ---------- |
| DINOv2-B, CLIP-B, SAM-B | ViT-S            | [BS_s1.pth](https://huggingface.co/yxlu0/SAK/blob/main/BS_s1.pth)  |
| DINOv2-B, CLIP-B, SAM-B | ViT-B            | [BB_s1.pth](https://huggingface.co/yxlu0/SAK/blob/main/BB_s1.pth)  |
| DINOv2-L, CLIP-L, SAM-L | ViT-B            | [LB_s1.pth](https://huggingface.co/yxlu0/SAK/blob/main/LB_s1.pth)  |
| DINOv2-L, CLIP-L, SAM-L | ViT-L            | [LL_s1.pth](https://huggingface.co/yxlu0/SAK/blob/main/LL_s1.pth)  |

**Stage 2**

We provide two example checkpoints after Stage 2 training, initialized by **BB_s1.pth** from Stage 1 training:

- PASCAL-Context: [BB_s2_pascal.pth](https://huggingface.co/yxlu0/SAK/blob/main/BB_s2_pascal.pth)
- NYUD-v2: [BB_s2_nyud.pth](https://huggingface.co/yxlu0/SAK/blob/main/BB_s2_nyud.pth)

## Citation

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

```bibtex
@inproceedings{lu2025swiss,
  title={Swiss Army Knife: Synergizing Biases in Knowledge from Vision Foundation Models for Multi-Task Learning},
  author={Yuxiang Lu and Shengcao Cao and Yu-Xiong Wang},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025}
}
```