---
license: apache-2.0
tags:
- computer-vision
- image-segmentation
- mask-generation
- vision-transformer
- onnx
- segment-anything
- sam
pipeline_tag: image-segmentation
---

# Segment Anything (SAM) ONNX Weights

This repository contains ONNX-optimized weights for the **Segment Anything Model (SAM)**, a foundation model designed for promptable and automatic mask generation (AMG).

The architecture is decoupled into separate **Image Encoder** and **Mask Decoder** ONNX graphs. This design allows running heavy image encoding once while executing lightweight mask decoding interactively across multiple point prompts on both CPU and GPU without PyTorch dependencies.

## Available Files
* **`vit_b_encoder.onnx` & `vit_b_decoder.onnx`**: ViT-Base backbone, optimized for reduced memory footprint and higher throughput.
* **`vit_l_encoder.onnx` & `vit_l_decoder.onnx`**: ViT-Large backbone, balanced representation capacity and inference speed.
* **`vit_h_encoder.onnx` & `vit_h_decoder.onnx`**: ViT-Huge backbone, standard high-fidelity segmentation model.

---

## How to Use

The easiest way to load and run these models is through the **[spatialhub](https://github.com/spatialhub-ai/spatialhub)** Python library:

```python
from spatialhub import SAM

# Initialize SAM automatic mask generator
segmentor = SAM(model_variant="vit_h")

# Generate instance mask proposals across a uniform grid
result = segmentor.generate_masks("image.jpg", points_per_side=32)

# Save segmentation overlay
result.visualize_mask("sam_output.png")
```


## Original Citation
If you use these models in academic work, please cite the original authors:
```bibtex
@article{kirillov2023segany,
  title={Segment Anything},
  author={Kirillov, Alexander and Mintun, Eric and Ravi, Nikhila and Mao, Hanzi and Rolland, Chloe and Gustafson, Laura and Xiao, Tete and Whitehead, Spencer and Berg, Alexander C. and Lo, Wan-Yen and Doll{\'a}r, Piotr and Girshick, Ross},
  journal={arXiv preprint arXiv:2304.02643},
  year={2023}
}
```