BiRefNet (fastforks)
BiRefNet (Zheng et al., CAAI AIR 2024): bilateral reference network for high-resolution dichotomous image segmentation and background removal. A Swin Transformer v1 encoder at full and half resolution feeds a decoder with deformable ASPP blocks, image patches as reference and gradient attention; one foreground mask per image. rmbg-2.0: BRIA's RMBG-2.0 weights for this network (non-commercial, gated).
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/ZhengPeng7/BiRefNet at f7079632d681.
Author weight repos: https://huggingface.co/ZhengPeng7/BiRefNet, https://huggingface.co/ZhengPeng7/BiRefNet_lite, https://huggingface.co/briaai/RMBG-2.0.
Task: matting. Tier: experimental. Outputs: alpha. Backends: portable, kernels.
Input: one RGB image (trimap-guided models also a trimap: background / unknown / foreground). Output: an alpha matte [H, W] in [0, 1] at the input size: 1 = foreground (background removal, dichotomous / salient object segmentation).
Variants
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|---|---|---|---|---|
birefnet@1 |
birefnet.safetensors | 9ab37426bf4d |
MIT | yes |
birefnet-lite@1 |
birefnet_lite.safetensors | 4417d8979525 |
MIT | yes |
rmbg-2.0@1 |
rmbg-2.0.safetensors | 566ed80c3d95 |
CC-BY-NC-4.0 | no |
Licence
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: MIT. Licence files: LICENSE.
Training data: DIS5K (dichotomous image segmentation), DUTS (salient object detection), HRSOD (high-resolution salient object detection), UHRSD (ultra-high-resolution salient object detection), HRS10K (high-resolution salient object detection), P3M-10k (privacy-preserving portrait matting) (General-use checkpoints: DIS5K-TR, DIS-TEs, DUTS-TR/TE, HRSOD-TR/TE, UHRSD-TR/TE, HRS10K-TR/TE, TR-P3M-10k, TE-P3M-500-NP, TE-P3M-500-P, TR-humans (HF model card). Some of these sets carry research-only terms; only the weights licence (MIT) counts for the commercial flag.).
Licence evidence: https://huggingface.co/ZhengPeng7/BiRefNet/blob/e2bf8e4460fc8fa32bba5ea4d94b3233d367b0e4/README.md ; https://huggingface.co/ZhengPeng7/BiRefNet_lite/blob/aa62cd87eafb9cc43056d08ef3615a14628b831d/README.md ; https://github.com/ZhengPeng7/BiRefNet/blob/f7079632d681ecf12161fcea8905682934e59e51/LICENSE
Input size
any H x W >= 32, padded to a multiple of 32
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): birefnet 3000x4000 (1.7 GiB peak allocated, 2.8 reserved); birefnet-lite 3000x4000 (0.9 GiB peak allocated, 1.8 reserved); rmbg-2.0 3000x4000 (1.7 GiB peak allocated, 2.6 reserved)
Speed and memory
Matting, synthetic, 1080x1920
| model | time (ms): original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|---|---|---|---|---|---|---|---|---|
birefnet |
234 | 131 | 121 | 1.9x | 1.1x | 1666 / 1930 | 1660 / 1908 | 1674 / 1916 |
birefnet-lite |
75.1 | 58.0 | 48.1 | 1.6x | 1.2x | 913 / 974 | 810 / 952 | 754 / 960 |
rmbg-2.0 |
149 | 132 | 121 | 1.2x | 1.1x | 1672 / 1990 | 1660 / 1908 | 1674 / 1936 |
Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|---|---|---|---|---|---|---|
birefnet |
3.60 | 0.46 | 13.58 | 0.39 | 0.56 | 0.37 |
birefnet-lite |
8.04 | 0.67 | 0.20 | 1.10 | 0.42 | 0.47 |
rmbg-2.0 |
6.42 | 0.94 | 0.64 | 0.94 | 0.49 | 0.47 |
torch.compile and TensorRT
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
|---|---|---|---|---|---|---|---|---|---|---|---|
birefnet |
1080x1920 | 131 ms | 138 ms | 87.8 ms | 121 ms | 93.8 s | 201 s | 1660 / 1908 | 1874 / 1932 | 1674 / 1916 | 1753 / 1814 |
birefnet-lite |
1080x1920 | 58.0 ms | 50.2 ms | 49.0 ms | 48.1 ms | 75.1 s | 156 s | 810 / 952 | 1469 / 1552 | 754 / 960 | 874 / 934 |
rmbg-2.0 |
1080x1920 | 132 ms | 110 ms | 88.1 ms | 121 ms | 82.9 s | 219 s | 1660 / 1908 | 1874 / 1932 | 1674 / 1936 | 1723 / 1784 |
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
Accuracy
On public datasets
Each image at its own size against its binary mask: MAE of the alpha, max F-beta (beta^2 = 0.3) over 256 thresholds (BiRefNet's evaluation). Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | kernels | note |
|---|---|---|---|---|---|---|
birefnet |
DIS5K validation (DIS-VD), all 470 images, MAE / max F-beta | 0.0325 / 0.9074 | ||||
birefnet |
DIS5K validation (DIS-VD), every 2nd image (235), MAE / max F-beta | 0.0338 / 0.9037 | 0.0338 / 0.9037 | 0.0338 / 0.9037 | 0.0338 / 0.9037 | |
birefnet-lite |
DIS5K validation (DIS-VD), every 2nd image (235), MAE / max F-beta | 0.0446 / 0.8620 | 0.0446 / 0.8619 | 0.0446 / 0.8620 | 0.0446 / 0.8620 | |
rmbg-2.0 |
DIS5K validation (DIS-VD), every 2nd image (235), MAE / max F-beta | 0.0205 / 0.9324 | 0.0206 / 0.9324 | 0.0205 / 0.9326 | 0.0205 / 0.9326 |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 45b1fe90 / a49aaba7, 2026-11-12 to 2026-11-14.
Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 7.8e-05 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 6e-06 and p99 <= 2.1e-05 (absolute error of the output alpha).
Usage
from fastforks import matting
model = matting.load("birefnet") # mode="auto": the preferred backend from recorded measurements
result = model("photo.jpg")
result.alpha # [H, W] float32 in [0, 1], input resolution
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: birefnet, birefnet-lite, rmbg-2.0.
Kernels backend
matting.load("birefnet", mode="kernels") runs our CUDA kernels (deform_conv). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.
Model tree for fastforks/birefnet
Base model
ZhengPeng7/BiRefNet