deit-b16-food101 / README.md
mnjm's picture
Upload folder using huggingface_hub
844452c verified
|
Raw
History Blame
809 Bytes
---
library_name: pytorch
pipeline_tag: image-classification
datasets:
- food101
tags:
- food101
- image-classification
- vision-transformer
- deit
license: mit
---
# DeiT-B-16 on Food-101
Raw training checkpoint exported from this repository:
https://github.com/mnjm/vit
## Files
- `DeiT-B-16-food101.pt`: raw checkpoint from this training run
- `config/`: Hydra config snapshot saved with the run
## Checkpoint
- Path: `DeiT-B-16-food101.pt`
- Dataset: `food101`
- Model: `DeiT-B/16`
- Teacher: `hf_hub:anonauthors/food101-resnet50`
## Metrics
| Split | Loss | Acc@1 | Acc@5 |
| --- | ---: | ---: | ---: |
| Train | `0.4401` | `98.16%` | `99.73%` |
| Val | `1.9847` | `62.92%` | `84.57%` |
## Usage
This is a raw checkpoint dump. To use it check [here](https://github.com/mnjm/vit/compute_acc.py)