Instructions to use 1999xia/ViT_Fast with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- timm
How to use 1999xia/ViT_Fast with timm:
import timm model = timm.create_model("hf_hub:1999xia/ViT_Fast", pretrained=True) - Notebooks
- Google Colab
- Kaggle
File size: 1,336 Bytes
2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb 2b1d31d 54ee1eb | 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 | ---
license: mit
library_name: timm
---
# ViT_Fast — Model Weights
This repository contains model checkpoints for the **Miss Patch** project.
**Code and documentation**: [github.com/DoraemonXia/ViT_Fast](https://github.com/DoraemonXia/ViT_Fast)
## Available Weights
### Baselines (ViT-B/16, IN-21K pretrained)
| File | Dataset | Test Acc |
|:-----|:-------|:--------:|
| `checkpoints/cifar100_vit_b16_ft/best_model.pth` | CIFAR-100 | 91.69% |
| `checkpoints/oxford_pets_vit_b16_ft/best_model.pth` | Oxford Pets | 93.32% |
| `checkpoints/food101_vit_b16_in21k/best_model.pth` | Food-101 | 91.37% |
### Downsampled (168x168)
| File | Dataset | Test Acc |
|:-----|:-------|:--------:|
| `checkpoints/cifar100_vit_b16_img168/best_model.pth` | CIFAR-100 | 91.56% |
| `checkpoints/oxford_pets_vit_b16_img168/best_model.pth` | Oxford Pets | 90.65% |
### MAE + Router
| File | Dataset | Keep Ratio | Test Acc |
|:-----|:-------|:---------:|:--------:|
| `checkpoints/cifar100_mae_patchsel_b16_keep75_distill/best_model.pth` | CIFAR-100 | 75% | 91.10% |
| `checkpoints/cifar100_mae_patchsel_b16_keep50/best_model.pth` | CIFAR-100 | 50% | 89.07% |
### Distilled Routers
| File | Dataset |
|:-----|:-------|
| `checkpoints/router_distill_cifar100/router.pth` | CIFAR-100 |
| `checkpoints/router_distill_food101/router.pth` | Food-101 |
|