---
license: mit
library_name: onnx
pipeline_tag: image-to-image
tags:
  - onnx
  - image-inpainting
  - object-removal
  - onnxruntime-web
---

# MI-GAN (ONNX pipeline)

MI-GAN, a fast GAN for image inpainting, as an ONNX pipeline export that crops round the hole, fills at 512 × 512 and pastes the result back at any size. [Card Anvil](https://github.com/Card-Anvil) uses it in the browser to remove watermarks from card art.

This is an unchanged copy of one file from [`andraniksargsyan/migan`](https://huggingface.co/andraniksargsyan/migan) at commit [`406830d`](https://huggingface.co/andraniksargsyan/migan/tree/406830d0fa60666da0071c342ad2fbc8f30c5c64), kept here so the app loads a file that can't change under it.

| File | Source path |
| --- | --- |
| `migan_pipeline_v2.onnx` | `migan_pipeline_v2.onnx` |

## Credits and licence

- Model: [MI-GAN](https://github.com/Picsart-AI-Research/MI-GAN) by Picsart AI Research (PAIR), from *MI-GAN: A Simple Baseline for Image Inpainting on Mobile Devices* (ICCV 2023).
- ONNX export: [`andraniksargsyan/migan`](https://huggingface.co/andraniksargsyan/migan).

Licensed under the MIT License, as the source is. See [`LICENSE`](LICENSE).