CoEx (fastforks)

CoEx (Bangunharcana et al. 2021): MobileNetV2 features, an attention-weighted correlation volume, 3-D aggregation with guided cost excitation, top-2 soft-argmax and superpixel upsampling (max disparity 192).

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/antabangun/coex at a98b37d33f37.

Task: stereo. Tier: stable. 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
coex-sceneflow@1	sceneflow.ckpt	e618be2d75cd	GPL-3.0-only (assumed from code)	yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: GPL-3.0-only. Files: LICENSE.

Code licence: GPL-3.0-only; the authors do not state a separate weights licence.

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

• Licence note: The code is GPL-3.0 (copyleft) and ships in the separate fastforks-nc distribution (pip install 'fastforks-core[nc]'); the authors state no weights licence.

Licence evidence: https://github.com/antabangun/coex/blob/a98b37d33f37754582b1e71f38216af8c69f2347/LICENSE

Input size

any H x W >= 32, padded to a multiple of 32 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): coex-sceneflow 3000x4000 (8.5 GiB peak allocated, 10.7 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
coex-sceneflow	94.9	60.0	49.93	23.48	2.6x	2.1x	2098 / 2160	1136 / 1254	495 / 886

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
coex-sceneflow	–	0.16	0.09†	0.63 (375x1242)	0.06†	0.04†

Fixed-size speed and peak memory, all models

model	mode	640x480	1280x720	1920x1080	12 MP 4000x3000
coex-sceneflow	portable	11.2 ms / 225 / 312	28.5 ms / 671 / 890	70.0 ms / 1485 / 1916	554.7 ms / 8660 / 10964
coex-sceneflow	kernels	4.3 ms / 103 / 322	13.7 ms / 256 / 400	32.9 ms / 559 / 804	211.1 ms / 3190 / 4410

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	portable	kernels	note
coex-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	33.46 / 15.994	33.46 / 15.989	33.46 / 16.000
coex-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	32.24 / 7.418	32.32 / 7.427	32.31 / 7.421

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-10-27.

Parity with the original code

Every variant is checked against golden 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.

Limitations

• SceneFlow-only weights: zero-shot output on real images is poor, so this is mainly a latency reference.
• The original code's KITTI weights (in its git repository) are not included.

Usage

from fastforks import stereo

model = stereo.load("coex-sceneflow")  # portable PyTorch on the GPU if there is one, else the CPU
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: coex-sceneflow.

Kernels backend

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