fastforks/igev-plusplus

IGEV++ (fastforks)

IGEV++ (Xu et al. 2025): IGEV-Stereo with geometry encoding volumes for small, medium and large disparity ranges (48 bins each at 1/4 resolution, up to 768 px), aggregated separately and blended per pixel in the ConvGRU update, 32 iterations.

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/gangweix/IGEV-plusplus at 099dec979897.

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
igev-plusplus-eth3d@1	eth3d.pth	4595ac1c164b	MIT (assumed from code)	yes
igev-plusplus-kitti12@1	kitti2012.pth	2e7947101eec	MIT (assumed from code)	yes
igev-plusplus-kitti15@1	kitti2015.pth	42db237df5f8	MIT (assumed from code)	yes
igev-plusplus-middlebury@1	middlebury.pth	eecddfa4549d	MIT (assumed from code)	yes
igev-plusplus-sceneflow@1	sceneflow.pth	550227fb7b73	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/gangweix/IGEV-plusplus/blob/099dec97989780d151700459c682df1cc2c18e88/README.md#L23

Input size

any H x W >= 32, padded to a multiple of 32 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): igev-plusplus-eth3d not measured yet; igev-plusplus-kitti12 not measured yet; igev-plusplus-kitti15 not measured yet; igev-plusplus-middlebury not measured yet; igev-plusplus-sceneflow not measured yet

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
igev-plusplus-eth3d	2012	1278	1081	393	3.3x	2.7x	2891 / 3108	1594 / 1840	1389 / 2302

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
igev-plusplus-eth3d	0.42 (375x1242)	0.20	0.16†	2.33 (375x1242)	1.10†	0.40†

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)
igev-plusplus-eth3d	994x1482	1081	715.6	159.0	494.4	425	393	1594 / 1840	952 / 1074	3632

Fixed-size speed and peak memory, all models

model	mode	480x640	720x1280	1080x1920	12 MP 3000x4000
igev-plusplus-eth3d	portable	201.5 ms / 354 / 614	606.5 ms / 987 / 1662	1534.8 ms / 2172 / 3496	OOM
igev-plusplus-eth3d	kernels	88.7 ms / 357 / 1092	241.2 ms / 830 / 1210	534.3 ms / 1799 / 2464	3135.8 ms (n=3) / 8419 / 9752
igev-plusplus-kitti12	portable	201.4 ms / 354 / 614	607.2 ms / 987 / 1662	1535.8 ms / 2172 / 3496	OOM
igev-plusplus-kitti12	kernels	89.1 ms / 357 / 1092	244.8 ms / 830 / 1210	535.9 ms / 1799 / 2464	3143.3 ms (n=3) / 8419 / 9750
igev-plusplus-kitti15	portable	202.7 ms / 354 / 614	606.9 ms / 987 / 1662	1535.8 ms / 2172 / 3496	OOM
igev-plusplus-kitti15	kernels	88.7 ms / 357 / 1092	240.2 ms / 830 / 1210	534.9 ms / 1799 / 2464	3141.8 ms (n=3) / 8419 / 9752
igev-plusplus-middlebury	portable	201.2 ms / 354 / 612	606.8 ms / 987 / 1662	1535.6 ms / 2172 / 3496	OOM
igev-plusplus-middlebury	kernels	88.7 ms / 357 / 1092	240.5 ms / 830 / 1210	535.0 ms / 1799 / 2464	3140.6 ms (n=3) / 8419 / 9752
igev-plusplus-sceneflow	portable	201.0 ms / 354 / 612	606.2 ms / 987 / 1662	1534.9 ms / 2172 / 3496	OOM
igev-plusplus-sceneflow	kernels	88.6 ms / 357 / 1092	240.6 ms / 830 / 1210	535.4 ms / 1799 / 2464	3143.9 ms (n=3) / 8419 / 9752

Apple silicon (MPS)

model	portable	kernels	speed-up
igev-plusplus-sceneflow	2852 ms / 1783 MiB (float16)	2054 ms / 1803 MiB	1.39x

Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).

• igev-plusplus-kitti12, igev-plusplus-kitti15, igev-plusplus-middlebury, igev-plusplus-sceneflow: same network and settings as igev-plusplus-eth3d, measured once (rows not repeated).

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
igev-plusplus-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	4.66 / 0.356	4.66 / 0.355	4.65 / 0.355	4.66 / 0.356
igev-plusplus-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	10.57 / 3.120	10.58 / 3.120	10.55 / 3.128	10.56 / 3.153
igev-plusplus-kitti12	ETH3D two-view train, bad > 1 px % / EPE	26.75 / 1.968	26.77 / 1.966	26.73 / 1.969	26.78 / 1.963
igev-plusplus-kitti12	Middlebury 2014 trainingH, bad > 2 px % / EPE	21.25 / 4.995	21.25 / 4.990	21.25 / 4.996	21.25 / 4.976
igev-plusplus-kitti15	ETH3D two-view train, bad > 1 px % / EPE	25.52 / 1.896	25.51 / 1.893	25.47 / 1.888	25.50 / 1.890
igev-plusplus-kitti15	Middlebury 2014 trainingH, bad > 2 px % / EPE	20.69 / 4.856	20.70 / 4.861	20.71 / 4.851	20.69 / 4.859
igev-plusplus-middlebury	ETH3D two-view train, bad > 1 px % / EPE	18.40 / 0.779	18.31 / 0.778	18.47 / 0.780	18.46 / 0.782	possibly in the training data
igev-plusplus-middlebury	Middlebury 2014 trainingH, bad > 2 px % / EPE	4.70 / 0.703	4.66 / 0.703	4.76 / 0.707	4.79 / 0.708	trained on this dataset's public images; not a zero-shot result
igev-plusplus-eth3d	ETH3D two-view train, bad > 1 px % / EPE	0.50 / 0.093	0.50 / 0.094	0.50 / 0.093	0.50 / 0.094	trained on this dataset's public images; not a zero-shot result
igev-plusplus-eth3d	Middlebury 2014 trainingH, bad > 2 px % / EPE	10.61 / 2.709	10.62 / 2.649	10.58 / 2.713	10.60 / 2.734

Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-05; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-06.

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.

Limitations

• CUDA output (float16, kernels, float32 with PyTorch's default TF32 convolutions) differs from the CPU by ~0.06 px mean / 1.2 px p99 at 720x1280, 0.005 px with full float32: rounding amplified by 32 GRU iterations. Portable and kernels reach the same ETH3D / Middlebury accuracy.

Usage

from fastforks import stereo

model = stereo.load("igev-plusplus-eth3d")  # 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: igev-plusplus-eth3d, igev-plusplus-kitti12, igev-plusplus-kitti15, igev-plusplus-middlebury, igev-plusplus-sceneflow.

Kernels backend

stereo.load("igev-plusplus-eth3d", mode="kernels") runs our CUDA kernels (igev, igevpp, 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.