ONNX Background Remover Model (RMBG-2.0)

This repository contains the ONNX export of the RMBG-2.0 background removal model.

Files

• model.onnx → Original FP32 model (≈1.02 GB)
• model_int8.onnx → Quantized INT8 model (smaller, faster)
• quantize.py → Script to quantize FP32 → INT8

Usage

Load Quantized Model

from huggingface_hub import hf_hub_download
import onnxruntime as ort

# Download quantized model
model_path = hf_hub_download("NDPDevloper/onnx", "model_int8.onnx")

# Create ONNXRuntime session
session = ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])