Cas-GwcNet (fastforks)

Cas-GwcNet (cascade stereo, Gu et al. 2020): two-stage cascade cost volume (48 then 24 disparity hypotheses, the second narrowed around the first result) on GwcNet-style 3-D hourglasses; group-wise correlation + concatenation (gc) or concatenation (c) volumes.

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/alibaba/cascade-stereo at 2ae7da65515c.

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
cas-gwcnet-c-sceneflow@1	casgwcnet-c-2.ckpt	9d30ee767111	MIT (assumed from code)	yes
cas-gwcnet-gc-sceneflow@1	casgwcnet.ckpt	8a3638501cd5	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/alibaba/cascade-stereo/blob/2ae7da65515c82025383a0fbee8d6fe063dc7101/CasStereoNet/README.md#L17-L19

Input size

• cas-gwcnet-c-sceneflow: any H x W >= 225, padded to a multiple of 32
• cas-gwcnet-gc-sceneflow: any H x W >= 32, padded to a multiple of 32 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): cas-gwcnet-c-sceneflow 1988x2880 (8.2 GiB peak allocated, 9.2 reserved); cas-gwcnet-gc-sceneflow 1988x2880 (9.4 GiB peak allocated, 14.1 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
cas-gwcnet-c-sceneflow	656	389	327	225	1.7x	1.5x	2791 / 2970	2006 / 2474	1725 / 1930
cas-gwcnet-gc-sceneflow	863	–	472	358	–	1.3x	–	2338 / 2794	2272 / 2514

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
cas-gwcnet-c-sceneflow	0.48 (375x1242)	1.29	0.18†	0.41 (375x1242)	0.33†	0.23†
cas-gwcnet-gc-sceneflow	0.25 (375x1242)	0.97	0.17†	0.25 (375x1242)	0.47†	0.36†

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)
cas-gwcnet-c-sceneflow	994x1482	327	294.6	61.7	220.3	131	225	2006 / 2474	2098 / 2448	1440

Fixed-size speed and peak memory, all models

model	mode	480x640	720x1280	1080x1920	12 MP 3000x4000
cas-gwcnet-c-sceneflow	portable	61.3 ms / 465 / 738	195.5 ms / 1386 / 2160	452.8 ms / 3046 / 4720	OOM
cas-gwcnet-c-sceneflow	kernels	44.8 ms / 442 / 568	135.5 ms / 1242 / 1586	316.1 ms / 2692 / 3408	OOM
cas-gwcnet-gc-sceneflow	portable	86.7 ms / 533 / 1166	281.6 ms / 1588 / 3308	647.4 ms / 3494 / 7262	OOM
cas-gwcnet-gc-sceneflow	kernels	66.9 ms / 510 / 956	213.7 ms / 1449 / 2606	493.8 ms / 3147 / 5634	OOM

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
cas-gwcnet-gc-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	8.08 / 0.821	99.90 / 13.127 (NaN / inf output)	8.08 / 0.823	8.09 / 0.823
cas-gwcnet-gc-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	23.83 / 6.541	99.80 / 70.838 (NaN / inf output)	23.84 / 6.547	23.83 / 6.541
cas-gwcnet-c-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	7.85 / 0.512	7.83 / 0.513	7.84 / 0.512	7.85 / 0.513
cas-gwcnet-c-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	30.07 / 8.568	29.90 / 8.490	30.07 / 8.569	30.07 / 8.574

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 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| <= 0.05 px and p99 <= 0.5 px.

Usage

from fastforks import stereo

model = stereo.load("cas-gwcnet-c-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: cas-gwcnet-c-sceneflow, cas-gwcnet-gc-sceneflow.

Kernels backend

stereo.load("cas-gwcnet-c-sceneflow", mode="kernels") runs our CUDA kernels (igev, 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.