OpenRemove：基于 BRIA RMBG-1.4 与 ONNX Runtime 的自托管离线 AI 抠图工具

A high-performance, self-hosted, offline AI background removal application and REST API powered by BRIA RMBG-1.4 and Microsoft ONNX Runtime.

Overview

Commercial background removal APIs often enforce restrictive quotas, usage tracking, recurring subscription models, or privacy concerns regarding cloud data transfers.

OpenRemove provides a self-hosted, private alternative designed for efficient local hardware execution. It isolates subjects and removes backgrounds using state-of-the-art dichotomous image segmentation in ~2-3 seconds on CPU with minimal memory footprint (<700MB peak RAM).

Key Features

BRIA RMBG-1.4 Segmentation: High-resolution image segmentation for fine edge preservation, hair matting, and foreground isolation.
Ultra-Lightweight & Fast: Slashes peak RAM to <700MB and completes inference in 2-3.3s on standard CPU cores.
Local Execution: Runs 100% offline via Microsoft ONNX Runtime (C++ engine) with AVX2 CPU acceleration.
Comparison Viewer: Built-in before/after split slider powered by CSS clipping.
Real-Time Progress Streaming: Server-Sent Events (SSE) provide live progress percentages and execution stages.
Clipboard & Drag-and-Drop: Supports direct image paste (Ctrl + V) and file selection.
REST API: Simple endpoints for integration with bot services, scripts, and automated workflows.

Directory Structure

openremove/
├── .github/
│   ├── ISSUE_TEMPLATE/
│   │   ├── bug_report.md
│   │   └── feature_request.md
│   └── PULL_REQUEST_TEMPLATE.md
├── docs/
│   ├── CODE_OF_CONDUCT.md
│   ├── CONTRIBUTING.md
│   └── SECURITY.md
├── models/
│   └── model.onnx              # RMBG-1.4 ONNX model weights (168 MB)
├── public/
│   ├── index.html              # Main background remover web interface
│   ├── how-to-use.html         # User guide page
│   ├── api-docs.html           # REST API reference documentation
│   ├── faq.html                # FAQ and technical limits
│   ├── privacy.html            # Privacy policy & data safety
│   ├── contributing.html       # Contribution guidelines
│   ├── status.html             # Real-time system status monitor
│   ├── robots.txt              # Crawler permissions
│   └── .well-known/
│       └── security.txt        # Vulnerability disclosure policy
├── tmp/                        # Temporary uploaded images & outputs
├── inference-worker.js         # Isolated worker for memory-safe inference
├── model-server.js             # Standalone Model Engine API service
├── server.js                   # Web gateway & REST API server
├── package.json                # Project dependencies and scripts
├── CHANGELOG.md                # Release history and updates
├── LICENSE                     # MIT License
└── README.md                   # Project documentation

Getting Started

Prerequisites

Node.js (version 18.0.0 or newer)
npm or yarn
ONNX Model weights (models/model.onnx)

1. Installation

Clone the repository and install dependencies:

git clone https://github.com/DDGXO/openremove.git
cd openremove
npm install

2. Model Weights

RMBG-1.4 (Pre-included): Included in the repository under models/model.onnx for out-of-the-box offline inference.

3. Environment Variables Configuration

Copy the template to create your local .env file:

cp .env.example .env

Variable	Default	Description
`PORT`	`3000`	Port for the web gateway interface and REST API
`BACKEND_URL`	`http://localhost:5000`	URL or domain of the backend model engine
`MODEL_PATH`	`./models/model.onnx`	Custom path to ONNX model weights

4. Running the Server

Option A: Standalone Mode (All-in-One)

Run the web application and local AI inference pipeline together on a single server:

npm start

Web Interface: http://localhost:3000
Healthcheck: http://localhost:3000/ping

Option B: Decoupled Mode (Microservice Architecture)

Run the Model Inference Engine on a dedicated compute machine and the Web Gateway on a separate node:

Start Model Engine Server (Port 5000 on compute machine):
```
npm run start:engine
```
Start Web Gateway (pointing to Backend Model Server):
```
BACKEND_URL=http://localhost:5000 npm run start:web
```

API Reference

Healthcheck

Check service status and uptime.

GET /ping
GET /api/ping

Response (200 OK):

{
  "status": true,
  "message": "pong",
  "app": "OpenRemove",
  "timestamp": 1789722300000
}

Upload Image

Upload an image for background removal.

POST /api/upload
Content-Type: multipart/form-data

Form Data:

image: Image file (JPG, PNG, WEBP, JFIF, max 25MB)

Response (200 OK):

{
  "status": true,
  "jobId": "1789722300-abc123xyz",
  "filename": "upload-1789722300-123456789.jpg"
}

Stream Inference & Progress (SSE)

Connect to the Server-Sent Events stream to receive real-time execution progress and download URLs.

GET /api/process-stream/:jobId

Events:

event: progress

{
  "percent": 40,
  "statusText": "Executing RMBG AI background segmentation on CPU...",
  "log": "[17:50:12] ONNX InferenceSession running on CPU (AVX2 mode)..."
}

event: complete

{
  "status": true,
  "durationMs": 2850,
  "originalUrl": "/tmp/upload-1789722300-123456789.jpg",
  "resultUrl": "/tmp/nobg-upload-1789722300-123456789.png",
  "downloadName": "nobg-upload-1789722300-123456789.png"
}

Contributing

Contributions are welcome. Please read CONTRIBUTING.md and our Code of Conduct before submitting pull requests or issues.

Acknowledgments

OpenRemove builds upon the work of the following open-source projects:

BRIA RMBG-1.4: State-of-the-art background removal model by BRIA AI.
Microsoft ONNX Runtime: Cross-platform inference engine.
Sharp & libvips: High-speed Node.js image processing library.
Hugging Face Community: Open-source model weights hosting and conversions.

License

This project is licensed under the MIT License.