Models

Stereo models

71 stereo models. Each section says what the model is, whether it may be used commercially and how fast it is; licences and checks are under Details.

Input: a rectified left / right image pair. Output: a disparity map (pixels, left view).

Measured on RTX 5060 Ti, torch 2.14.1+cu130; fastforks bench MODEL --size HxW measures yours. The summary uses one input, 994x1482: kernels time with its speed-up over the faster original run, and kernels peak reserved memory with its ratio to the smaller original one. Where measured (one variant per family), a variant's table also lists portable under torch.compile and the fp16 TensorRT engine of portable (export; memory: the engine's device memory). – = not measured. The accuracy column is lower-is-better (ETH3D bad > 1 px (%)).

Other devices: Apple M1 Pro 16 GB, NVIDIA L4, NVIDIA L40S, NVIDIA A10, NVIDIA A100-SXM4-40GB, NVIDIA RTX PRO 6000; across devices.

modelcommercialETH3D bad > 1 px (%) ↓kernels time (ms),
speed-up vs original		kernels peak memory (MiB, reserved),
ratio to original
aanet-plus-sceneflowyes18.5088.91, 2.2x920, 0.24x
aanet-sceneflowyes14.64100.81, 1.8x644, 0.57x
bridgedepth-rvc-pretrainno1.22 ?353.63, 1.3x2686, 0.85x
bridgedepth-sfno1.24348.37, 1.3x2634, 0.84x
c-fast-foundationstereo (max_disp 192, valid_iters 8)yes2.20134.91, 3.9x1516, 0.51x
c-foundationstereo-syes1.241362.1, 2.8x4376, 0.34x
cas-gwcnet-c-sceneflowyes7.85225.4, 1.7x1930, 0.65x
cas-gwcnet-gc-sceneflowyes8.09357.52, 1.6x2514, 0.29x
cfnet-sceneflowyes5.60400.67, 1.4x4718, 0.6x
coex-sceneflowyes33.4623.48, 2.6x886, 0.41x
crestereo-eth3dyes1.39 †228.95, 7.0x732, 0.42x
defom-stereo-vitl-sceneflowno2.44658.71, 4.3x3566, 0.41x
dlnr-middleburyyes34.38213.87, 2.8x1606, 0.45x
dlnr-sceneflowyes30.27213.87, 2.8x1606, 0.45x
fast-acvnet-generalizationyes6.5138.16, 1.6x1042, 0.69x
fast-acvnet-kitti12yes19.0638.16, 1.6x1042, 0.69x
fast-acvnet-kitti15yes17.5938.16, 1.6x1042, 0.69x
fast-acvnet-plus-generalizationyes6.2046.41, 1.7x1120, 0.74x
fast-acvnet-plus-kitti12yes21.6246.41, 1.7x1120, 0.74x
fast-acvnet-plus-kitti15yes21.6346.41, 1.7x1120, 0.74x
fast-acvnet-plus-sceneflowyes12.6246.41, 1.7x1120, 0.74x
fast-acvnet-sceneflowyes8.4338.16, 1.6x1042, 0.69x
fast-foundationstereo (max_disp 192, valid_iters 8)no1.72134.98, 3.9x1576, 0.48x
foundationstereo-vit-lno0.65 ?1513.07, 2.8x3300, 0.38x
foundationstereo-vit-sno1.10 ?1120.17, 3.2x2470, 0.34x
great-igev-eth3dyes0.49 †393.14, 3.7x1564, 0.17x
great-igev-kittiyes23.36393.14, 3.7x1564, 0.17x
great-igev-middlebury (max_disp 768)yes27.53 ?578.811744
great-igev-sceneflowyes4.44393.14, 3.7x1564, 0.17x
gwcnet-g-sceneflowyes9.59238.13, 1.5x2102, 0.39x
gwcnet-gc-sceneflowyes10.96257.33, 1.4x1962, 0.32x
hitnet-eth3dyes0.93 †23.16, 7.9x1248
hitnet-middlebury-d400yes7.0752.73, 9.2x2368
igev-plusplus-eth3dyes0.50 †393.04, 3.3x2302, 0.74x
igev-plusplus-kitti12yes26.78393.04, 3.3x2302, 0.74x
igev-plusplus-kitti15yes25.50393.04, 3.3x2302, 0.74x
igev-plusplus-middleburyyes18.46 ?393.04, 3.3x2302, 0.74x
igev-plusplus-sceneflowyes4.66393.04, 3.3x2302, 0.74x
igev-stereo-eth3dyes0.56 ?288.25, 3.3x1374, 0.71x
igev-stereo-kitti12yes26.02288.25, 3.3x1374, 0.71x
igev-stereo-kitti15yes24.63288.25, 3.3x1374, 0.71x
igev-stereo-middleburyyes19.80 ?288.25, 3.3x1374, 0.71x
igev-stereo-sceneflowyes4.04288.25, 3.3x1374, 0.71x
las2-hyes2.14 ?53.27, 2.5x1388, 1.3x
las2-lyes2.24 ?29.92, 3.0x600, 0.55x
las2-myes2.98 ?18.29, 2.9x572, 0.57x
las2-syes4.10 ?14.35, 3.0x558, 0.57x
mocha-v1-sceneflowyes4.35455.46, 3.2x1730, 0.43x
mocha-v2-sceneflowyes3.82448.12, 2.9x1790, 0.45x
monster-mix-allno1.25 †667.63, 6.4x2714, 0.42x
monster-plusplus-mix-allno1.08 ?697.92, 6.2x2760, 0.43x
monster-sceneflowno1.34667.63, 6.4x2714, 0.42x
nmrf-sceneflowyes3.81195.66, 1.6x1112, 0.54x
psmnet-sceneflowyes15.13231.81, 2.6x2122, 0.35x
raft-stereo-eth3dyes2.61 ?272.78, 3.8x802, 0.31x
raft-stereo-middleburyyes2.79272.78, 3.8x802, 0.31x
raft-stereo-sceneflowyes2.60272.78, 3.8x802, 0.31x
rt-monster-plusplus-mix-allno2.05 ?94.13, 2.7x3976, 1.3x
s2m2-lno0.82 ?375.83, 2.7x1568, 0.38x
s2m2-mno0.95 ?247.87, 2.7x1162, 0.4x
s2m2-sno1.10 ?113.45, 3.1x820, 0.41x
s2m2-xlno0.52 ?681.94, 2.6x2274, 0.37x
selective-igev-eth3d-finetuneyes0.32 †348.76, 4.2x1390, 0.65x
selective-igev-eth3d-trainyes0.78 †348.76, 4.2x1390, 0.65x
selective-igev-kitti12yes24.07348.76, 4.2x1390, 0.65x
selective-igev-kitti15yes24.26348.76, 4.2x1390, 0.65x
selective-igev-middlebury-finetuneyes24.24348.76, 4.2x1390, 0.65x
selective-igev-middlebury-trainyes10.25348.76, 4.2x1390, 0.65x
selective-igev-sceneflowyes6.06348.76, 4.2x1390, 0.65x
waft-stereo-bno0.74 ?392.88, 2.9x2316, 0.39x
waft-stereo-syes0.95 ?283.76, 2.4x1286, 0.38x

† trained or fine-tuned on this dataset's public images; not a zero-shot result.

? training data not stated; overlap with this dataset's public images cannot be excluded.

All models

Modes, licences, training data and outputs per model

Commercial status per variant: yes if the code and the weights licence both allow commercial use (the training data is listed but does not change it; weights without a licence of their own are assumed to carry the code licence); no if either forbids it; unclear for custom licence terms (treat as no).

modelmodescode licenceweights licencetraining datacommercialoutputs
aanet-plus-sceneflowportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflowyesdisparity
aanet-sceneflowportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflowyesdisparity
bridgedepth-rvc-pretrainportable, kernelsMITCC-BY-NC-4.0sceneflow, falling_things, crestereo, tartanair, sintel, unstatedno (weights: CC-BY-NC-4.0)disparity
bridgedepth-sfportable, kernelsMITCC-BY-NC-4.0sceneflowno (weights: CC-BY-NC-4.0)disparity
c-fast-foundationstereoportable, kernelsApache-2.0LicenseRef-NVIDIA-Open-Model-Agreementfsd, nvidia-internal, stereo4dyesdisparity
c-foundationstereo-sportable, kernelsApache-2.0LicenseRef-NVIDIA-Open-Model-Licensefsd, crestereo, tartanairyesdisparity
cas-gwcnet-c-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
cas-gwcnet-gc-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
cfnet-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
coex-sceneflowportable, kernelsGPL-3.0-onlyGPL-3.0-only (assumed from code)sceneflowyesdisparity
crestereo-eth3dportable, kernelsApache-2.0Apache-2.0 (assumed from code)crestereo, sceneflow, sintel, middlebury, eth3d, kitti, falling_things, instereo2k, hr_vsyesdisparity
defom-stereo-vitl-sceneflowportable, kernelsMITCC-BY-NC-4.0sceneflowno (weights: CC-BY-NC-4.0)disparity
dlnr-middleburyportable, kernelsMITMIT (assumed from code)sceneflow, middleburyyesdisparity
dlnr-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
fast-acvnet-generalizationportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
fast-acvnet-kitti12portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
fast-acvnet-kitti15portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
fast-acvnet-plus-generalizationportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
fast-acvnet-plus-kitti12portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
fast-acvnet-plus-kitti15portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
fast-acvnet-plus-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
fast-acvnet-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
fast-foundationstereoportable, kernelsApache-2.0LicenseRef-NVIDIA-Source-Code-License-NCfsdno (weights: LicenseRef-NVIDIA-Source-Code-License-NC)disparity
foundationstereo-vit-lportable, kernelsLicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0LicenseRef-NVIDIA-Source-Code-License-NCunstatedno (code: LicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0 (noncommercial); weights: LicenseRef-NVIDIA-Source-Code-License-NC)disparity
foundationstereo-vit-sportable, kernelsLicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0LicenseRef-NVIDIA-Source-Code-License-NCunstatedno (code: LicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0 (noncommercial); weights: LicenseRef-NVIDIA-Source-Code-License-NC)disparity
great-igev-eth3dportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflow, crestereo, instereo2k, eth3d, unstatedyesdisparity
great-igev-kittiportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflow, kittiyesdisparity
great-igev-middleburyportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflow, crestereo, instereo2k, hr_vs, middlebury, falling_things, unstatedyesdisparity
great-igev-sceneflowportable, kernelsApache-2.0Apache-2.0 (assumed from code)sceneflowyesdisparity
gwcnet-g-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
gwcnet-gc-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
hitnet-eth3dportable, kernelsApache-2.0Apache-2.0 (assumed from code)eth3dyesdisparity
hitnet-middlebury-d400portable, kernelsApache-2.0Apache-2.0 (assumed from code)middleburyyesdisparity
igev-plusplus-eth3dportable, kernelsMITMIT (assumed from code)sceneflow, tartanair, sintel, crestereo, instereo2k, eth3dyesdisparity
igev-plusplus-kitti12portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
igev-plusplus-kitti15portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
igev-plusplus-middleburyportable, kernelsMITMIT (assumed from code)sceneflow, tartanair, falling_things, crestereo, instereo2k, middlebury, unstatedyesdisparity
igev-plusplus-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
igev-stereo-eth3dportable, kernelsMITMIT (assumed from code)unstatedyesdisparity
igev-stereo-kitti12portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
igev-stereo-kitti15portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
igev-stereo-middleburyportable, kernelsMITMIT (assumed from code)unstatedyesdisparity
igev-stereo-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
las2-hportable, kernelsMITMITunstatedyesdisparity
las2-lportable, kernelsMITMITunstatedyesdisparity
las2-mportable, kernelsMITMITunstatedyesdisparity
las2-sportable, kernelsMITMITunstatedyesdisparity
mocha-v1-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
mocha-v2-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
monster-mix-allportable, kernelsLicenseRef-noncommercialLicenseRef-noncommercial AND CC-BY-NC-4.0sceneflow, kitti, eth3d, middlebury, tartanair, crestereo, falling_things, instereo2k, sintel, hr_vsno (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial AND CC-BY-NC-4.0)disparity
monster-plusplus-mix-allportable, kernelsLicenseRef-noncommercialLicenseRef-noncommercial AND CC-BY-NC-4.0unstatedno (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial AND CC-BY-NC-4.0)disparity
monster-sceneflowportable, kernelsLicenseRef-noncommercialLicenseRef-noncommercial AND CC-BY-NC-4.0sceneflowno (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial AND CC-BY-NC-4.0)disparity
nmrf-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
psmnet-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
raft-stereo-eth3dportable, kernelsMITMIT (assumed from code)unstatedyesdisparity
raft-stereo-middleburyportable, kernelsMITMIT (assumed from code)sceneflow, middleburyyesdisparity
raft-stereo-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
rt-monster-plusplus-mix-allportable, kernelsLicenseRef-noncommercialLicenseRef-noncommercialunstatedno (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial)disparity
s2m2-lportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)disparity, confidence, occlusion
s2m2-mportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)disparity, confidence, occlusion
s2m2-sportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)disparity, confidence, occlusion
s2m2-xlportable, kernelsCC-BY-NC-4.0CC-BY-NC-4.0unstatedno (code: CC-BY-NC-4.0 (noncommercial); weights: CC-BY-NC-4.0)disparity, confidence, occlusion
selective-igev-eth3d-finetuneportable, kernelsMITMIT (assumed from code)sceneflow, eth3dyesdisparity
selective-igev-eth3d-trainportable, kernelsMITMIT (assumed from code)sceneflow, eth3dyesdisparity
selective-igev-kitti12portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
selective-igev-kitti15portable, kernelsMITMIT (assumed from code)sceneflow, kittiyesdisparity
selective-igev-middlebury-finetuneportable, kernelsMITMIT (assumed from code)sceneflow, middleburyyesdisparity
selective-igev-middlebury-trainportable, kernelsMITMIT (assumed from code)sceneflow, middleburyyesdisparity
selective-igev-sceneflowportable, kernelsMITMIT (assumed from code)sceneflowyesdisparity
waft-stereo-bportable, kernelsBSD-3-ClauseCC-BY-NC-4.0sceneflow, unstatedno (weights: CC-BY-NC-4.0)disparity
waft-stereo-sportable, kernelsBSD-3-ClauseBSD-3-Clausesceneflow, unstatedyesdisparity

Training datasets

iddatasetlicencecommercial use
sceneflowFreiburg SceneFlow (FlyingThings3D, Driving, Monkaa)LicenseRef-research-onlyno
kittiKITTI stereo 2012 / 2015CC-BY-NC-SA-3.0no
eth3dETH3D two-view stereoCC-BY-NC-SA-4.0no
middleburyMiddlebury stereo (2014)not statedunclear
crestereoCREStereo synthetic stereo dataset (Megvii)not statedunclear
sintelMPI Sintel stereonot statedunclear
falling_thingsNVIDIA Falling Things (FAT)CC-BY-NC-SA-4.0no
instereo2kInStereo2KLicenseRef-noncommercialno
hr_vsHR-VS (high-resolution virtual stereo, Yang et al. CVPR 2019)not statedunclear
fsdFoundation Stereo Dataset (FSD, NVIDIA synthetic)LicenseRef-NVIDIA-Source-Code-License-NCno
tartanairTartanAirCC-BY-4.0yes
stereo4dStereo4D (stereo YouTube videos)CC-BY-4.0unclear
nvidia-internalNVIDIA internal proprietary datanot statedunclear