BridgeDepth (fastforks)

BridgeDepth (Guan et al. 2025): Depth Anything V2 (ViT-L) features combined with NMRF stereo hypotheses; window-attention message passing between mono and stereo embeddings, inference at 1/8 resolution plus a 1/4 refinement.

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/aeolusguan/BridgeDepth at 9a231601e698.

Author weight repos: https://huggingface.co/aeolusguan/BridgeDepth.

Task: stereo. Tier: experimental. Outputs: disparity. Backends: portable, kernels.

Input: a rectified left / right image pair. Output: a disparity map (pixels, left view).

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
bridgedepth-rvc-pretrain@1	bridge_rvc_pretrain.pth	4cb8d0395e67	CC-BY-NC-4.0	no
bridgedepth-sf@1	bridge_sf.pth	d2cd44b6f749	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. Files: LICENSE, LICENSE-torchvision.

Training data: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa).

• Licence note: The weights embed Depth Anything V2 weights released under CC-BY-NC-4.0, so they are non-commercial even though the code is MIT.

Licence evidence: https://github.com/aeolusguan/BridgeDepth/blob/9a231601e6981d3536fa6a78ebaccebc1edb99a4/README.md#L40

Input size

any H x W >= 16, padded to a multiple of 16 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): bridgedepth-rvc-pretrain 3000x4000 (13.3 GiB peak allocated, 14.2 reserved); bridgedepth-sf 3000x4000 (13.3 GiB peak allocated, 14.2 reserved)

Speed and memory

Stereo, Middlebury 2014 half, 994x1482

model	time (ms): original fp32	original	portable	kernels	kernels speed-up: vs original	vs portable	peak memory (MiB, allocated / reserved): original	portable	kernels
bridgedepth-rvc-pretrain	629	470	442	354	1.3x	1.2x	2903 / 3200	2968 / 3246	2361 / 2686
bridgedepth-sf	616	460	437	348	1.3x	1.3x	2852 / 3144	2916 / 3214	2310 / 2634

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
bridgedepth-rvc-pretrain	5.87 (375x1242)	1.84	0.76†	1.36 (375x1242)	0.45†	0.36†
bridgedepth-sf	4.86 (375x1242)	1.53	0.72†	0.12 (375x1242)	0.44†	0.35†

torch.compile and TensorRT

model	input	portable ms	compile ms	compile s (cold)	TensorRT fp16 ms	ONNX + build s	kernels ms	peak MiB (alloc / reserved): portable	compile	TensorRT (device)
bridgedepth-sf	994x1482	437	325.1	111.4	fails: TensorRT bug: Myelin internal error	–	348	2916 / 3214	2280 / 2308	–

Fixed-size speed and peak memory, all models

model	mode	480x640	720x1280	1080x1920	12 MP 3000x4000
bridgedepth-rvc-pretrain	portable	76.7 ms / 1727 / 1798	247.2 ms / 2359 / 2628	676.5 ms / 3562 / 4102	7918.7 ms (n=3) / 13620 / 14590
bridgedepth-rvc-pretrain	kernels	62.2 ms / 1147 / 1266	200.9 ms / 1764 / 2096	534.7 ms / 2939 / 3514	5573.1 ms (n=3) / 12950 / 13592
bridgedepth-sf	portable	73.2 ms / 1676 / 1748	242.8 ms / 2308 / 2560	668.4 ms / 3511 / 4032	7849.6 ms (n=3) / 13569 / 14540
bridgedepth-sf	kernels	59.1 ms / 1096 / 1214	196.3 ms / 1714 / 2044	526.9 ms / 2888 / 3462	5498.6 ms (n=3) / 12898 / 13540

Apple silicon (MPS)

model	portable	kernels	speed-up
bridgedepth-sf	1065 ms / 4145 MiB (float32)	1012 ms / 4093 MiB	1.05x

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 authors' code 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

Original = the authors' code, same checkpoint and protocol.

variant	dataset	original fp32	original fp16	portable	kernels	note
bridgedepth-sf	ETH3D two-view train, bad > 1 px % / EPE	1.25 / 0.230	1.25 / 0.229	1.24 / 0.229	1.24 / 0.229
bridgedepth-sf	Middlebury 2014 trainingH, bad > 2 px % / EPE	9.97 / 2.111	9.99 / 2.109	10.00 / 2.111	10.00 / 2.109
bridgedepth-rvc-pretrain	ETH3D two-view train, bad > 1 px % / EPE	1.22 / 0.191	1.22 / 0.192	1.22 / 0.191	1.22 / 0.191	possibly in the training data
bridgedepth-rvc-pretrain	Middlebury 2014 trainingH, bad > 2 px % / EPE	6.35 / 0.950	6.35 / 0.948	6.38 / 0.953	6.37 / 0.952	possibly in the training data

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-05.

Parity with the original code

Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 0.002 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.02 px and p99 <= 0.25 px.

Usage

from fastforks import stereo

# non-commercial weights: run with FASTFORKS_NONCOMMERCIAL=1 in the environment
model = stereo.load("bridgedepth-rvc-pretrain")  # mode="auto": the preferred backend from recorded measurements
result = model("left.png", "right.png")
result.disparity  # [H, W] float32 tensor, 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: bridgedepth-rvc-pretrain, bridgedepth-sf.

Kernels backend

stereo.load("bridgedepth-rvc-pretrain", mode="kernels") runs our CUDA kernels (fastbb). 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.