ormbg-ONNX (Orangebot mirror)

ONNX weights of ormbg, an open background-removal model based on IS-Net, for running in the browser with Transformers.js.

This repo powers the free, in-browser Background Remover on orangebot.ai: photos are processed on the visitor's device and never uploaded.

Source and changes

• Files are copied unchanged from onnx-community/ormbg-ONNX at revision 034e2d884afbab897e10e78fc5bb566b29533fd6, which converts schirrmacher/ormbg to ONNX.
• Only two precisions are kept: onnx/model_fp16.onnx (WebGPU with shader-f16) and onnx/model_uint8.onnx (WebAssembly / CPU). Other quantized variants were left out.
• No weights were modified. Credit for the model goes to its author (schirrmacher); credit for the ONNX conversion goes to the ONNX Community.

Usage (Transformers.js)

import { pipeline } from '@huggingface/transformers';

const remover = await pipeline('background-removal', 'Orangebot/ormbg-ONNX', {
  device: 'webgpu', // or 'wasm'
  dtype: 'fp16',    // use 'uint8' with 'wasm'
});
const [result] = await remover('photo.jpg');
// result is a RawImage with the background made transparent

Limitations

The upstream author describes ormbg as optimized for images with humans. Products, pets, fine hair against busy backgrounds and transparent objects (glass) may give weaker cut-outs.

License

Apache License 2.0, the same as the upstream model; see LICENSE.