IS-Net general-use model (DIS, "Highly Accurate Dichotomous Image Segmentation", Xuebin Qin et al., ECCV 2022) Source: https://github.com/xuebinqin/DIS Licence: Apache License 2.0 ONNX export: rembg (https://github.com/danielgatis/rembg), MIT licence. Modifications: convolution weights stored as 8-bit integers per output channel (DequantizeLinear), maths unchanged in float32; the resize sizes read from the image so the model runs at any input size. The file is split into two parts for hosting; part0.bin followed by part1.bin is the complete ONNX model.