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.
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 |
| atc_base-im1k-single | ImageNet-1K | Single | link |
| atc_base-im1k-complete | ImageNet-1K | Complete | link |
| atc_small-im1k-average | ImageNet-1K | Average | link |
| atc_small-im1k-single | ImageNet-1K | Single | link |
| atc_small-im1k-complete | ImageNet-1K | Complete | link |
| atc_tiny-im1k-average | ImageNet-1K | Average | link |
| atc_tiny-im1k-single | ImageNet-1K | Single | link |
| atc_tiny-im1k-complete | ImageNet-1K | Complete | link |
| atc_base-nab-average | NABirds | Average | link |
| atc_base-nab-single | NABirds | Single | link |
| atc_base-nab-complete | NABirds | Complete | link |
| atc_small-nab-average | NABirds | Average | link |
| atc_small-nab-single | NABirds | Single | link |
| atc_small-nab-complete | NABirds | Complete | link |
| atc_tiny-nab-average | NABirds | Average | link |
| atc_tiny-nab-single | NABirds | Single | link |
| atc_tiny-nab-complete | NABirds | Complete | link |
| atc_base-coco-average | COCO 2014 | Average | link |
| atc_base-coco-single | COCO 2014 | Single | link |
| atc_base-coco-complete | COCO 2014 | Complete | link |
| atc_small-coco-average | COCO 2014 | Average | link |
| atc_small-coco-single | COCO 2014 | Single | link |
| atc_small-coco-complete | COCO 2014 | Complete | link |
| atc_tiny-coco-average | COCO 2014 | Average | link |
| atc_tiny-coco-single | COCO 2014 | Single | link |
| atc_tiny-coco-complete | COCO 2014 | Complete | link |
| atc_base-nus-average | NUS-WIDE | Average | link |
| atc_base-nus-single | NUS-WIDE | Single | link |
| atc_base-nus-complete | NUS-WIDE | Complete | link |
| atc_small-nus-average | NUS-WIDE | Average | link |
| atc_small-nus-single | NUS-WIDE | Single | link |
| atc_small-nus-complete | NUS-WIDE | Complete | link |
| atc_tiny-nus-average | NUS-WIDE | Average | link |
| atc_tiny-nus-single | NUS-WIDE | Single | link |
| atc_tiny-nus-complete | NUS-WIDE | Complete | link |
Object Detection and Segmentation Models:
| Model Name | Dataset | Keep Rate | Weights |
|---|---|---|---|
| atc_small-coco17-average | COCO 2017 | Average | link |
| atc_small-coco17-single | COCO 2017 | Single | link |
| atc_small-coco17-complete | COCO 2017 | Complete | link |
| tome_small-coco17 | COCO 2017 | - | link |
| atc_tiny-coco17-average | COCO 2017 | Average | link |
| atc_tiny-coco17-single | COCO 2017 | Single | link |
| atc_tiny-coco17-complete | COCO 2017 | Complete | link |
| tome_tiny-coco17 | COCO 2017 | - | link |
