File size: 6,807 Bytes
faa5063
 
 
 
 
 
 
2b7de45
faa5063
 
 
 
 
 
a18a1ea
 
 
 
faa5063
 
 
9b057aa
faa5063
 
 
 
 
019d5b1
faa5063
 
019d5b1
faa5063
4846471
 
 
 
0189224
4846471
 
0189224
4846471
faa5063
4846471
 
 
 
 
 
 
 
 
faa5063
4846471
 
 
 
 
 
 
 
 
faa5063
4846471
 
 
 
 
 
 
 
 
faa5063
 
9b057aa
faa5063
 
4846471
 
 
 
faa5063
4846471
 
 
 
faa5063
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
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
---
license: mit
language:
- en
---

# Model Details
Agglomerative Token Clustering (ATC), a novel hierarchical hard-merging based token reduction method.
ATC has been demonstrated to outperform all prior state-of-the-art token reduction methods across image classification, image synthesis, and object detection & segementation tasks.

We provide DeiT checkpoints (Tiny, Small, and Base) at four reduction rates (0.9, 0.7, 0.5, and 0.25) for four classification datasets: ImageNet-1K, NABirds, COCO 2014, and NUS-WIDE.
We also provide ViT-Adapter checkpoints (Tiny and Small) for the COCO 2017 object detection and segementation task, for both the ATC and ToMe token reduction methods.


### Model Description
- **Developed by:** Joakim Bruslund Haurum, Sergio Escalera, Graham W. Taylor, and Thomas B. Moeslund
- **Model type:** Vision Transformers
- **License:** MIT License

### More Resources
- **Repository:** [https://github.com/JoakimHaurum/ATC](https://github.com/JoakimHaurum/ATC)
- **Paper:** [https://arxiv.org/abs/2409.11923](https://arxiv.org/abs/2409.11923)
- **Project Page:** [https://vap.aau.dk/atc](https://vap.aau.dk/atc)
- **HuggingFace Collection:** [https://huggingface.co/collections/joakimbh/agglomerative-token-clustering-66e94dfb313e85ec97590fe4](https://huggingface.co/collections/joakimbh/agglomerative-token-clustering-66e94dfb313e85ec97590fe4)


## Model Zoo
**Note: This repository does not host any checkpoints but contains links to all the model repositories. Each token reduction method repository contains the checkpoints for the four considered keep rates.**

Image Classification Models:
| Model Name | Dataset | Linkage Function | Weights |
|:-----------|:--------:|:-----------:|:--------:|
| atc_base-im1k-average | ImageNet-1K | Average | [link](https://huggingface.co/joakimbh/atc_base-im1k-average) |
| atc_base-im1k-single | ImageNet-1K | Single | [link](https://huggingface.co/joakimbh/atc_base-im1k-single) |
| atc_base-im1k-complete | ImageNet-1K | Complete | [link](https://huggingface.co/joakimbh/atc_base-im1k-complete) |
| atc_small-im1k-average | ImageNet-1K | Average | [link](https://huggingface.co/joakimbh/atc_small-im1k-average) |
| atc_small-im1k-single | ImageNet-1K | Single | [link](https://huggingface.co/joakimbh/atc_small-im1k-single) |
| atc_small-im1k-complete | ImageNet-1K | Complete | [link](https://huggingface.co/joakimbh/atc_small-im1k-complete) |
| atc_tiny-im1k-average | ImageNet-1K | Average | [link](https://huggingface.co/joakimbh/atc_tiny-im1k-average) |
| atc_tiny-im1k-single | ImageNet-1K | Single | [link](https://huggingface.co/joakimbh/atc_tiny-im1k-single) |
| atc_tiny-im1k-complete | ImageNet-1K | Complete | [link](https://huggingface.co/joakimbh/atc_tiny-im1k-complete) |
<br>
| atc_base-nab-average | NABirds | Average | [link](https://huggingface.co/joakimbh/atc_base-nab-average) |
| atc_base-nab-single | NABirds | Single | [link](https://huggingface.co/joakimbh/atc_base-nab-single) |
| atc_base-nab-complete | NABirds | Complete | [link](https://huggingface.co/joakimbh/atc_base-nab-complete) |
| atc_small-nab-average | NABirds | Average | [link](https://huggingface.co/joakimbh/atc_small-nab-average) |
| atc_small-nab-single | NABirds | Single | [link](https://huggingface.co/joakimbh/atc_small-nab-single) |
| atc_small-nab-complete | NABirds | Complete | [link](https://huggingface.co/joakimbh/atc_small-nab-complete) |
| atc_tiny-nab-average | NABirds | Average | [link](https://huggingface.co/joakimbh/atc_tiny-nab-average) |
| atc_tiny-nab-single | NABirds | Single | [link](https://huggingface.co/joakimbh/atc_tiny-nab-single) |
| atc_tiny-nab-complete | NABirds | Complete | [link](https://huggingface.co/joakimbh/atc_tiny-nab-complete) |
<br>
| atc_base-coco-average | COCO 2014 | Average | [link](https://huggingface.co/joakimbh/atc_base-coco-average) |
| atc_base-coco-single | COCO 2014 | Single | [link](https://huggingface.co/joakimbh/atc_base-coco-single) |
| atc_base-coco-complete | COCO 2014 | Complete | [link](https://huggingface.co/joakimbh/atc_base-coco-complete) |
| atc_small-coco-average | COCO 2014 | Average | [link](https://huggingface.co/joakimbh/atc_small-coco-average) |
| atc_small-coco-single | COCO 2014 | Single | [link](https://huggingface.co/joakimbh/atc_small-coco-single) |
| atc_small-coco-complete | COCO 2014 | Complete | [link](https://huggingface.co/joakimbh/atc_small-coco-complete) |
| atc_tiny-coco-average | COCO 2014 | Average | [link](https://huggingface.co/joakimbh/atc_tiny-coco-average) |
| atc_tiny-coco-single | COCO 2014 | Single | [link](https://huggingface.co/joakimbh/atc_tiny-coco-single) |
| atc_tiny-coco-complete | COCO 2014 | Complete | [link](https://huggingface.co/joakimbh/atc_tiny-coco-complete) |
<br>
| atc_base-nus-average | NUS-WIDE | Average | [link](https://huggingface.co/joakimbh/atc_base-nus-average) |
| atc_base-nus-single | NUS-WIDE | Single | [link](https://huggingface.co/joakimbh/atc_base-nus-single) |
| atc_base-nus-complete | NUS-WIDE | Complete | [link](https://huggingface.co/joakimbh/atc_base-nus-complete) |
| atc_small-nus-average | NUS-WIDE | Average | [link](https://huggingface.co/joakimbh/atc_small-nus-average) |
| atc_small-nus-single | NUS-WIDE | Single | [link](https://huggingface.co/joakimbh/atc_small-nus-single) |
| atc_small-nus-complete | NUS-WIDE | Complete | [link](https://huggingface.co/joakimbh/atc_small-nus-complete) |
| atc_tiny-nus-average| NUS-WIDE | Average | [link](https://huggingface.co/joakimbh/atc_tiny-nus-average) |
| atc_tiny-nus-single | NUS-WIDE | Single | [link](https://huggingface.co/joakimbh/atc_tiny-nus-single) |
| atc_tiny-nus-complete | NUS-WIDE | Complete | [link](https://huggingface.co/joakimbh/atc_tiny-nus-complete) |
<br>

Object Detection and Segmentation Models:
| Model Name | Dataset | Keep Rate | Weights |
|:-----------|:--------:|:-----------:|:--------:|
| atc_small-coco17-average | COCO 2017 | Average | [link](https://huggingface.co/joakimbh/atc_small-coco17-average) |
| atc_small-coco17-single | COCO 2017 | Single | [link](https://huggingface.co/joakimbh/atc_small-coco17-single) |
| atc_small-coco17-complete | COCO 2017 | Complete | [link](https://huggingface.co/joakimbh/atc_small-coco17-complete) |
| tome_small-coco17 | COCO 2017 | - | [link](https://huggingface.co/joakimbh/tome_small-coco17) |
<br>
| atc_tiny-coco17-average | COCO 2017 | Average | [link](https://huggingface.co/joakimbh/atc_tiny-coco17-average) |
| atc_tiny-coco17-single | COCO 2017 | Single | [link](https://huggingface.co/joakimbh/atc_tiny-coco17-single) |
| atc_tiny-coco17-complete | COCO 2017 | Complete | [link](https://huggingface.co/joakimbh/atc_tiny-coco17-complete) |
| tome_tiny-coco17 | COCO 2017 | - | [link](https://huggingface.co/joakimbh/tome_tiny-coco17) |
<br>