Skip to content

Model weights v1 — BiRefNet_lite family (Core ML)

Choose a tag to compare

View all tags
@RuochenLyu RuochenLyu released this 28 Jul 07:28
· 60 commits to main since this release
models-v1
5ac8f66

Core ML conversions of the BiRefNet_lite background-removal models, downloaded on demand by Pluck (never bundled with the app).

Asset Model Size SHA256
BiRefNetLite.mlpackage.zip BiRefNet_lite 82.6 MB 4614141d1e3ce84bd5713b995beabbb225e6e568522b296c4c7f48279228e817
BiRefNetLiteMatting.mlpackage.zip BiRefNet_lite-matting 82.5 MB 7ea2236f7494ab7ac46b070847c00768cfd940249af62dd6ac1a367ed222f22a

Licence: MIT, Copyright (c) 2024 ZhengPeng (BiRefNet). Each archive carries the upstream LICENSE and a NOTICE with full attribution.

Provenance: converted from the original fp32 safetensors by Scripts/convert-birefnet.py (fp16, ImageNet normalization folded into the input layer, deform_conv2d rewritten in plain tensor ops). Numerical parity against the fp32 originals — MAE ≤ 0.0005, binary-mask agreement ≥ 99.96% on real photos — is recorded in docs/research.md appendix A.6.

Pluck verifies the pinned SHA256 of every download before installing (see product-plan §4.8); these hashes are compiled into the app's manifest.

Assets 4
Loading