remove-background

Fast, minimal background removal for Python using ONNX segmentation models.

Features

• Uses ONNX Runtime for fast inference (CPU)
• Minimal dependencies: only onnxruntime, Pillow, numpy
• Core image preprocessing and postprocessing written from scratch
• Python API and CLI
• Works with any ONNX segmentation model (ISNet, U2-Net, etc.)

Installation

pip install remove-background

The import name is remove_background:

Usage

Python API

from remove_background import remove_background, get_mask

# Remove background (returns RGBA image)
result = remove_background("input.jpg", model_path="model.onnx")
result.save("output.png")

# Get mask only (returns grayscale image)
mask = get_mask("input.jpg", model_path="model.onnx")
mask.save("mask.png")

CLI

# Remove background
python -m remove_background input.jpg -o output.png -m model.onnx

# Get mask only
python -m remove_background input.jpg --mask-only -m model.onnx

Model Setup

Download an ONNX segmentation model (e.g., ISNet or U2-Net) and place it in your working directory, or pass model_path explicitly.

The model must:

• Accept input shape (1, 3, H, W) float32
• Output a segmentation mask (1, 1, H, W) or (1, H, W)

License

MIT