---
license: mit
tags:
  - chess
  - object-detection
  - yolo
  - fen
  - ocr
library_name: ultralytics
---

# SmallVision — 2D Chess OCR Models

Canonical **smallchess** org repo for finetuned YOLO and RF-DETR weights used by [SmallVision](https://github.com/smallchess/SmallVision).

Detects chessboards, pieces, orientation markers, and last-move tiles in 2D chess screenshots.

Derived from [AndrewSpano/2d-chess-ocr](https://huggingface.co/AndrewSpano/2d-chess-ocr).

> Previously mirrored from the user repo `scchess/smallvision`. Use **`smallchess/smallvision`** going forward.

## Files

| File | Backend | Size |
|------|---------|------|
| `yolo26m-finetuned.pt` | YOLO | ~42 MB |
| `yolo26m-finetuned.onnx` | YOLO ONNX | ~78 MB |
| `yolo26n-finetuned.pt` | YOLO | ~5 MB |
| `yolo26n-finetuned.onnx` | YOLO ONNX | ~10 MB |
| `rfdetr-l-finetuned.pth` | RF-DETR-L | ~129 MB |
| `rfdetr-2xl-finetuned.pth` | RF-DETR-2XL | ~484 MB |

## Usage

```python
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="smallchess/smallvision",
    filename="yolo26m-finetuned.pt",
    repo_type="model",
)
```

CLI:

```bash
hf download smallchess/smallvision --repo-type model --local-dir ./finetuned-models
```

See [SmallVision](https://github.com/smallchess/SmallVision) for the full inference workflow.