AANet (fastforks)

AANet (Xu and Zhang 2020): sparse deformable-sampling cost aggregation within and across scales on a 2-D correlation pyramid, without 3-D convolutions.

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/haofeixu/aanet at 8d0d747e69bb.

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
aanet-plus-sceneflow@1	aanet+_sceneflow.pth	d3e13ef00d3d	Apache-2.0 (assumed from code)	yes
aanet-sceneflow@1	aanet_sceneflow.pth	5aa5a24e2ba8	Apache-2.0 (assumed from code)	yes

Licence

Original-code strings, recorded as they are (research-use package; check them before any other use). Code: Apache-2.0. Files: LICENSE.

Code licence: Apache-2.0; the authors do not state a separate weights licence.

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

Licence evidence: https://github.com/haofeixu/aanet/blob/8d0d747e69bbdba43e2578b60ad7df63ba7fdac2/LICENSE

Input size

• aanet-plus-sceneflow: any H x W >= 96, padded to a multiple of 96
• aanet-sceneflow: any H x W >= 48, padded to a multiple of 48 Tested up to (portable, fp16, RTX 5060 Ti 16 GB): aanet-plus-sceneflow 1024x2048 (7.1 GiB peak allocated, 8.1 reserved); aanet-sceneflow 3000x4000 (9.4 GiB peak allocated, 12.5 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
aanet-plus-sceneflow	262.9	194.2	159.22	88.91	2.2x	1.8x	3652 / 3784	1142 / 1278	813 / 920
aanet-sceneflow	289.2	178.0	139.79	100.81	1.8x	1.4x	1000 / 1126	845 / 986	536 / 644

Startup and first call

model	load (s): original	portable	kernels	first call (s): original	portable	kernels
aanet-plus-sceneflow	0.15 (375x1242)	0.16	0.20†	0.70 (375x1242)	0.17†	0.09†
aanet-sceneflow	0.12 (375x1242)	0.19	0.27†	0.07 (375x1242)	0.14†	0.10†

Fixed-size speed and peak memory, all models

model	mode	640x480	1280x720	1920x1080	12 MP 4000x3000
aanet-plus-sceneflow	portable	58.4 ms / 1109 / 1292	170.2 ms / 3400 / 3858	389.0 ms / 7214 / 8226	SKIPPED
aanet-plus-sceneflow	kernels	28.8 ms / 178 / 244	53.9 ms / 501 / 696	128.5 ms / 1049 / 1430	793.2 ms / 5752 / 8230
aanet-sceneflow	portable	40.1 ms / 300 / 340	94.2 ms / 781 / 830	231.2 ms / 1711 / 1820	1591.3 ms / 9603 / 12836
aanet-sceneflow	kernels	25.9 ms / 148 / 196	59.2 ms / 344 / 602	152.4 ms / 728 / 1252	967.9 ms / 3967 / 6854

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
aanet-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	14.64 / 0.697	14.63 / 0.697	14.64 / 0.698
aanet-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	32.15 / 6.305	32.16 / 6.301	32.16 / 6.303
aanet-plus-sceneflow	ETH3D two-view train, bad > 1 px % / EPE	18.48 / 0.811	18.50 / 0.811	18.51 / 0.812
aanet-plus-sceneflow	Middlebury 2014 trainingH, bad > 2 px % / EPE	29.84 / 6.237	29.86 / 6.237	29.85 / 6.232

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

• Only the SceneFlow checkpoints of AANet and AANet+ are included; the original code's KITTI fine-tunes are not.
• At 375x1242 AANet is launch-bound: portable is about as fast as the original code. At 994x1482 portable is faster but allocates more memory.

Usage

from fastforks import stereo

model = stereo.load("aanet-plus-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: aanet-plus-sceneflow, aanet-sceneflow.

Kernels backend

stereo.load("aanet-plus-sceneflow", mode="kernels") runs our CUDA kernels (deform_conv, 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.