MoCha-Stereo (fastforks)

MoCha-Stereo (Chen et al. 2024): EfficientNetV2-L features, a motif channel correlation volume next to an IGEV-style geometry volume, 3-level Mogrifier LSTM updates and a refinement step, 32 iterations. mocha-v2-sceneflow is an unofficial reconstruction (the authors released no V2 code).

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/ZYangChen/MoCha-Stereo at 78c5889c467f.

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
mocha-v1-sceneflow@1	mocha-stereo.pth	4d26f51d6321	MIT (assumed from code)	yes
mocha-v2-sceneflow@1	mocha-V2.pth	b7cd8969158b	MIT (assumed from code)	yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: MIT. Files: LICENSE.

Code licence: MIT; the authors do not state a separate weights licence.

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

Licence evidence: https://github.com/ZYangChen/MoCha-Stereo/blob/78c5889c467fe31414c9425f1280afc0a8f6c1a0/README.md#L69

Input size

any H x W >= 32, padded to a multiple of 32 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): mocha-v1-sceneflow 3000x4000 (12.2 GiB peak allocated, 15.2 reserved); mocha-v2-sceneflow 3000x4000 (12.2 GiB peak allocated, 14.8 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
mocha-v1-sceneflow	2148	1456	1057	455	3.2x	2.3x	3637 / 3982	1632 / 1770	1588 / 1730
mocha-v2-sceneflow	2054	1295	1033	448	2.9x	2.3x	3576 / 3982	1632 / 1770	1587 / 1790

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
mocha-v1-sceneflow	1.95 (375x1242)	0.79	0.48†	1.85 (375x1242)	1.06†	0.45†
mocha-v2-sceneflow	2.37 (375x1242)	1.30	0.45†	0.40 (375x1242)	1.03†	0.45†

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)
mocha-v2-sceneflow	994x1482	1033	712.6	148.1	437.0	361	448	1632 / 1770	1122 / 1308	1098

Fixed-size speed and peak memory, all models

model	mode	480x640	720x1280	1080x1920	12 MP 3000x4000
mocha-v1-sceneflow	portable	231.1 ms / 500 / 662	628.3 ms / 1078 / 1438	1484.8 ms / 2164 / 2688	10361.4 ms (n=3) / 12466 / 15520
mocha-v1-sceneflow	kernels	135.4 ms / 495 / 694	317.0 ms / 1055 / 1338	718.5 ms / 2092 / 2484	4396.8 ms (n=3) / 11664 / 14958
mocha-v2-sceneflow	portable	227.0 ms / 500 / 662	613.9 ms / 1078 / 1438	1451.9 ms / 2164 / 2688	10135.5 ms (n=3) / 12464 / 15164
mocha-v2-sceneflow	kernels	133.6 ms / 494 / 654	312.9 ms / 1054 / 1456	709.3 ms / 2091 / 2482	4316.6 ms (n=3) / 11664 / 14404

Apple silicon (MPS)

model	portable	kernels	speed-up
mocha-v1-sceneflow	2734 ms / 2403 MiB (float16)	2277 ms / 2382 MiB	1.20x

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
mocha-v1-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	4.35 / 1.087	4.35 / 1.084	4.35 / 1.085	4.35 / 1.088
mocha-v1-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	9.56 / 2.584	9.57 / 2.582	9.57 / 2.588	9.56 / 2.588
mocha-v2-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	3.84 / 0.892	3.83 / 0.885	3.83 / 0.887	3.82 / 0.886
mocha-v2-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	10.76 / 2.838	10.78 / 2.843	10.73 / 2.844	10.77 / 2.833

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

model = stereo.load("mocha-v1-sceneflow")  # 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: mocha-v1-sceneflow, mocha-v2-sceneflow.

Kernels backend

stereo.load("mocha-v1-sceneflow", mode="kernels") runs our CUDA kernels (igev, mocha, cutlass_conv, cudnn_hacc). 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.