Models

174 models, 142 commercially usable.

Times are medians of warm runs on the RTX 5060 Ti 16 GB, the reference GPU, at each task's standard input: portable is the model in plain PyTorch, kernels the same weights with the custom kernels. This GPU also has the authors' original code, torch.compile and TensorRT: switch on the extra columns. Click a model for its variants, accuracy and limitations; the method is on the benchmarks page. Data from fastforks a55e747e.

Device

Task

Commercially usable only Extra columns: original code, torch.compile, TensorRT, accuracy 174 shown

		Time (ms)		Kernels speed-up vs	Peak memory (MiB)		Licence (code / weights)
							Licence (code / weights)
`squeezenet1_1`	Classification	1.8	0.4	4.5×	41	8	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`mobilenetv1_100`	Classification	1.8	0.4	4.5×	44	76	commercial Apache-2.0 / Apache-2.0
`mobilenet_v2`	Classification	3.0	0.4	7.5×	43	75	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`resnet18`	Classification	1.5	0.5	3.0×	61	90	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`mobilenet_v3_large`	Classification	3.8	0.5	7.6×	46	78	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`alexnet`	Classification	1.0	0.6	1.7×	151	184	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`mobilenetv4_conv_medium`	Classification	3.9	0.6	6.5×	54	87	commercial Apache-2.0 / Apache-2.0
`efficientnet_b0`	Classification	5.0	0.8	6.3×	48	78	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`shufflenet_v2_x1_0`	Classification	3.7	0.8	4.6×	41	87	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`resnet50`	Classification	3.1	0.8	3.9×	89	116	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`dinov2-vits14-reg`	Feature backbones	3.0	0.8	3.8×	80	112	commercial Apache-2.0 / Apache-2.0
`dinov2-vits14`	Feature backbones	3.0	0.8	3.8×	79	112	commercial Apache-2.0 / Apache-2.0
`dinov3-vits16`	Feature backbones	4.7	0.9	5.2×	77	77	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`efficientvit_b2`	Classification	11.9	1.1	10.8×	85	120	commercial Apache-2.0 / Apache-2.0
`dinov3-convnext-tiny`	Feature backbones	3.0	1.2	2.5×	93	86	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`densenet121`	Classification	9.4	1.3	7.2×	53	83	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`inception_v3`	Classification	5.2	1.3	4.0×	120	150	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`fastvit_sa24`	Classification	6.9	1.4	4.9×	81	112	commercial Apache-2.0 / LicenseRef-Apple-Sample-Code
`dinov3-vitb16`	Feature backbones	4.9	1.4	3.5×	201	201	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`vgg16`	Classification	2.0	1.5	1.3×	313	332	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`dinov2-vitb14`	Feature backbones	2.9	1.5	1.9×	205	237	commercial Apache-2.0 / Apache-2.0
`dinov2-vitb14-reg`	Feature backbones	2.9	1.5	1.9×	207	237	commercial Apache-2.0 / Apache-2.0
`deit_base_patch16_224`	Classification	3.0	1.6	1.9×	203	300	commercial Apache-2.0 / Apache-2.0
`vit_base_patch16_224`	Classification	3.0	1.6	1.9×	205	300	commercial Apache-2.0 / Apache-2.0
`resnet152`	Classification	8.4	1.8	4.7×	155	182	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`regnety_160`	Classification	8.3	1.9	4.4×	208	229	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`efficientnet_v2_s`	Classification	10.2	1.9	5.4×	81	114	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`efficientnet_b4`	Classification	10.0	2.0	5.0×	89	111	commercial BSD-3-Clause / BSD-3-Clause (assumed from code)
`dinov3-convnext-small`	Feature backbones	5.4	2.0	2.7×	134	127	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`swin_base_patch4_window7_224`	Classification	10.8	2.1	5.1×	213	246	commercial Apache-2.0 / MIT
`dinov3-convnext-base`	Feature backbones	5.7	2.2	2.6×	208	200	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`dfine-n-coco`	Object detection	10.3	2.7	3.8×	88	124	commercial Apache-2.0 / Apache-2.0
`convnextv2_base`	Classification	10.2	2.9	3.5×	299	318	non-commercial Apache-2.0 / CC-BY-NC-4.0
`dinov3-vitl16`	Feature backbones	9.1	3.3	2.8×	617	617	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`segformer-b0-ade20k`	Semantic segmentation	11.4	3.5	3.3×	339	148	non-commercial Apache-2.0 / LicenseRef-NVIDIA-Source-Code-License-NC
`dinov3-convnext-large`	Feature backbones	5.4	3.6	1.5×	419	441	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`dfine-s-coco`	Object detection	11.1	4.0	2.8×	105	138	commercial Apache-2.0 / Apache-2.0
`rtdetrv2-s-coco`	Object detection	8.0	4.2	1.9×	134	155	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`depth-anything-v2-s`	Monocular depth	7.8	4.5	1.7×	249	190	commercial Apache-2.0 / Apache-2.0
`dinov2-vitl14`	Feature backbones	6.1	4.6	1.3×	622	655	commercial Apache-2.0 / Apache-2.0
`dinov2-vitl14-reg`	Feature backbones	6.2	4.6	1.3×	625	655	commercial Apache-2.0 / Apache-2.0
`dinov2_vitl14_lc`	Classification	5.4	4.9	1.1×	623	890	commercial Apache-2.0 / Apache-2.0
`rtdetrv2-m-coco`	Object detection	10.4	5.3	2.0×	156	177	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`rtdetrv2-ms-coco`	Object detection	9.6	5.5	1.7×	166	189	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`dfine-m-coco`	Object detection	13.5	6.0	2.3×	126	156	commercial Apache-2.0 / Apache-2.0
`rtdetrv2-l-coco`	Object detection	12.1	6.6	1.8×	177	200	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`dfine-l-coco`	Object detection	16.2	8.7	1.9×	152	179	commercial Apache-2.0 / Apache-2.0
`rtdetrv2-x-coco`	Object detection	14.6	9.2	1.6×	244	262	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`dinov3-vith16plus`	Feature backbones	13.3	9.3	1.4×	1645	1644	commercial LicenseRef-DINOv3-License / LicenseRef-DINOv3-License
`detr-r50-coco`	Object detection	12.4	9.9	1.3×	280	218	commercial Apache-2.0 / Apache-2.0
`depth-anything-v2-b`	Monocular depth	18.8	10.5	1.8×	450	387	non-commercial Apache-2.0 / CC-BY-NC-4.0
`hrnet-w48-fcn-ade20k`	Semantic segmentation	26.9	11.9	2.3×	424	274	commercial Apache-2.0 / Apache-2.0
`dinov2-s-linear-ade20k`	Semantic segmentation	18.3	12.4	1.5×	642	645	commercial Apache-2.0 / Apache-2.0
`vit_so400m_patch14_siglip_378`	Classification	13.1	12.8	1.0×	879	1216	commercial Apache-2.0 / Apache-2.0
`dinov2-vitg14-reg`	Feature backbones	18.7	13.0	1.4×	2217	2222	commercial Apache-2.0 / Apache-2.0
`dinov2-vitg14`	Feature backbones	18.9	13.2	1.4×	2216	2246	commercial Apache-2.0 / Apache-2.0
`detr-r101-coco`	Object detection	18.7	13.7	1.4×	316	254	commercial Apache-2.0 / Apache-2.0
`sam2.1-hiera-tiny`	Promptable segmentation	16.6	13.9	1.2×	304	316	commercial Apache-2.0 / Apache-2.0
`eva02_large_patch14_448`	Classification	18.0	14.1	1.3×	648	1236	commercial Apache-2.0 / MIT
`dfine-x-coco`	Object detection	18.5	14.1	1.3×	227	237	commercial Apache-2.0 / Apache-2.0
`las2-s`	Stereo matching	46.9	14.3	3.3×	955	320	commercial MIT / MIT
`da3mono-large`	Monocular depth	23.9	14.6	1.6×	818	814	commercial Apache-2.0 / Apache-2.0
`da3metric-large`	Monocular depth	24.1	14.7	1.6×	815	811	commercial Apache-2.0 / Apache-2.0
`sam2.1-hiera-small`	Promptable segmentation	18.8	15.6	1.2×	318	330	commercial Apache-2.0 / Apache-2.0
`fcn-r101-d8-ade20k`	Semantic segmentation	40.7	16.6	2.5×	417	294	commercial Apache-2.0 / Apache-2.0
`da3-large-1.1`	Monocular depth	37.4	17.1	2.2×	992	985	non-commercial Apache-2.0 / CC-BY-NC-4.0
`deeplabv3plus-r101-d8-ade20k`	Semantic segmentation	47.1	18.0	2.6×	413	365	commercial Apache-2.0 / Apache-2.0
`las2-m`	Stereo matching	54.1	18.3	3.0×	972	329	commercial MIT / MIT
`faster-rcnn-r50-fpn-3x-coco`	Object detection	22.8	19.5	1.2×	390	391	commercial Apache-2.0 / CC-BY-SA-3.0
`deeplabv3-r101-d8-ade20k`	Semantic segmentation	47.9	20.1	2.4×	453	362	commercial Apache-2.0 / Apache-2.0
`mask-rcnn-r50-fpn-3x-coco`	Object detection	24.6	21.0	1.2×	352	353	commercial Apache-2.0 / CC-BY-SA-3.0
`dinov2-b-linear-ade20k`	Semantic segmentation	33.3	22.6	1.5×	765	771	commercial Apache-2.0 / Apache-2.0
`hitnet-eth3d`	Stereo matching	67.8	23.2	2.9×	860	647	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`coex-sceneflow`	Stereo matching	49.9	23.5	2.1×	1136	495	commercial GPL-3.0-only / GPL-3.0-only (assumed from code)
`pspnet-r101-d8-ade20k`	Semantic segmentation	48.8	25.0	2.0×	417	341	commercial Apache-2.0 / Apache-2.0
`sam2.1-hiera-base-plus`	Promptable segmentation	32.0	25.1	1.3×	412	426	commercial Apache-2.0 / Apache-2.0
`segformer-b5-ade20k`	Semantic segmentation	38.7	25.3	1.5×	546	407	non-commercial Apache-2.0 / LicenseRef-NVIDIA-Source-Code-License-NC
`patchmatchnet`	Multi-view depth	202	29.6	6.8×	1806	740	commercial MIT / MIT (assumed from code)
`moge-2-vits-normal`	Monocular depth	64.2	29.7	2.2×	1066	884	commercial MIT / MIT
`las2-l`	Stereo matching	80.3	29.9	2.7×	1036	397	commercial MIT / MIT
`depth-anything-v2-l`	Monocular depth	55.1	31.1	1.8×	1026	921	non-commercial Apache-2.0 / CC-BY-NC-4.0
`eomt-l-dinov2-ade20k`	Semantic segmentation	50.8	35.0	1.5×	1515	1515	commercial Apache-2.0 / MIT
`mvsanywhere`	Multi-view depth	93.2	35.6	2.6×	931	761	non-commercial LicenseRef-noncommercial AND CC-BY-NC-SA-4.0 / LicenseRef-noncommercial AND CC-BY-NC-SA-4.0 (assumed from code)
`sam-vit-b`	Promptable segmentation	48.3	37.0	1.3×	378	377	commercial Apache-2.0 / Apache-2.0
`fast-acvnet-generalization`	Stereo matching	54.7	38.2	1.4×	664	739	commercial MIT / MIT (assumed from code)
`fast-acvnet-kitti12`	Stereo matching	54.7	38.2	1.4×	664	739	commercial MIT / MIT (assumed from code)
`fast-acvnet-kitti15`	Stereo matching	54.7	38.2	1.4×	664	739	commercial MIT / MIT (assumed from code)
`fast-acvnet-sceneflow`	Stereo matching	54.7	38.2	1.4×	664	739	commercial MIT / MIT (assumed from code)
`eomt-l-dinov3-ade20k`	Semantic segmentation	54.6	38.8	1.4×	1514	1514	commercial Apache-2.0 / MIT AND LicenseRef-DINOv3-License
`da3-giant-1.1`	Monocular depth	81.8	44.2	1.9×	2638	2628	non-commercial Apache-2.0 / CC-BY-NC-4.0
`fast-acvnet-plus-generalization`	Stereo matching	70.6	46.4	1.5×	662	790	commercial MIT / MIT (assumed from code)
`fast-acvnet-plus-kitti12`	Stereo matching	70.6	46.4	1.5×	662	790	commercial MIT / MIT (assumed from code)
`fast-acvnet-plus-kitti15`	Stereo matching	70.6	46.4	1.5×	662	790	commercial MIT / MIT (assumed from code)
`fast-acvnet-plus-sceneflow`	Stereo matching	70.6	46.4	1.5×	662	790	commercial MIT / MIT (assumed from code)
`sam2.1-hiera-large`	Promptable segmentation	68.0	48.8	1.4×	742	760	commercial Apache-2.0 / Apache-2.0
`hitnet-middlebury-d400`	Stereo matching	176	52.7	3.3×	1744	1263	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`las2-h`	Stereo matching	123	53.3	2.3×	1050	1021	commercial MIT / MIT
`dinov2-l-linear-ade20k`	Semantic segmentation	83.9	56.8	1.5×	1181	1189	commercial Apache-2.0 / Apache-2.0
`da3nested-giant-large-1.1`	Monocular depth	108	58.2	1.9×	3275	3271	non-commercial Apache-2.0 / CC-BY-NC-4.0
`mask2former-swinl-ade20k`	Semantic segmentation	103	60.4	1.7×	1164	1164	non-commercial Apache-2.0 / CC-BY-NC-4.0
`unik3d-vitl`	Monocular depth	108	62.9	1.7×	1219	1166	non-commercial CC-BY-NC-SA-4.0 / CC-BY-NC-SA-4.0 (assumed from code)
`midas-dpt-beit-large-512`	Monocular depth	78.8	63.9	1.2×	992	1003	commercial MIT / MIT (assumed from code)
`unidepth-v2-vitl14`	Monocular depth	99.8	66.4	1.5×	1208	1211	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0 (assumed from code)
`moge-2-vitl-normal`	Monocular depth	145	73.4	2.0×	1661	1498	commercial MIT / MIT
`sam-vit-l`	Promptable segmentation	116	81.5	1.4×	833	831	commercial Apache-2.0 / Apache-2.0
`aanet-plus-sceneflow`	Stereo matching	159	88.9	1.8×	1142	813	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`zoedepth-nk`	Monocular depth	124	88.9	1.4×	2037	1771	commercial MIT / MIT (assumed from code)
`sapiens-seg-0.3b-goliath`	Semantic segmentation	132	92.6	1.4×	1643	1643	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`rt-monster-plusplus-mix-all`	Stereo matching	113	94.1	1.2×	2592	2704	non-commercial LicenseRef-noncommercial / LicenseRef-noncommercial
`oneformer-swinl-ade20k`	Semantic segmentation	139	100	1.4×	2054	2054	commercial Apache-2.0 / MIT
`aanet-sceneflow`	Stereo matching	140	101	1.4×	845	536	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`s2m2-s`	Stereo matching	348	114	3.1×	1733	686	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`sam3`	Promptable segmentation	179	117	1.5×	1922	2019	commercial LicenseRef-SAM-License AND Apache-2.0 / LicenseRef-SAM-License
`c-fast-foundationstereo`	Stereo matching	330	135	2.4×	2109	1196	commercial Apache-2.0 / LicenseRef-NVIDIA-Open-Model-Agreement
`fast-foundationstereo`	Stereo matching	330	135	2.4×	2109	1196	non-commercial Apache-2.0 / LicenseRef-NVIDIA-Source-Code-License-NC
`sam-vit-h`	Promptable segmentation	215	143	1.5×	1484	1480	commercial Apache-2.0 / Apache-2.0
`metric3d-vit-large`	Monocular depth	200	159	1.3×	1484	1516	commercial BSD-2-Clause / BSD-2-Clause (assumed from code)
`sapiens-seg-1b-goliath`	Semantic segmentation	275	185	1.5×	3259	3259	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`nmrf-sceneflow`	Stereo matching	216	196	1.1×	1510	1022	commercial MIT / MIT (assumed from code)
`dlnr-middlebury`	Stereo matching	398	214	1.9×	2635	1236	commercial MIT / MIT (assumed from code)
`dlnr-sceneflow`	Stereo matching	398	214	1.9×	2635	1236	commercial MIT / MIT (assumed from code)
`moge-3-vitl`	Monocular depth	514	215	2.4×	1686	1673	commercial MIT / MIT
`cas-gwcnet-c-sceneflow`	Stereo matching	327	225	1.5×	2006	1725	commercial MIT / MIT (assumed from code)
`crestereo-eth3d`	Stereo matching	732	229	3.2×	884	644	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`psmnet-sceneflow`	Stereo matching	340	232	1.5×	2078	1786	commercial MIT / MIT (assumed from code)
`gwcnet-g-sceneflow`	Stereo matching	344	238	1.4×	2081	1799	commercial MIT / MIT (assumed from code)
`s2m2-m`	Stereo matching	670	248	2.7×	2300	876	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`gwcnet-gc-sceneflow`	Stereo matching	353	257	1.4×	2082	1768	commercial MIT / MIT (assumed from code)
`depth-pro`	Monocular depth	518	271	1.9×	3306	3667	non-commercial LicenseRef-Apple-Sample-Code / LicenseRef-research-only
`raft-stereo-eth3d`	Stereo matching	746	273	2.7×	1388	765	commercial MIT / MIT (assumed from code)
`raft-stereo-middlebury`	Stereo matching	746	273	2.7×	1388	765	commercial MIT / MIT (assumed from code)
`raft-stereo-sceneflow`	Stereo matching	746	273	2.7×	1388	765	commercial MIT / MIT (assumed from code)
`waft-stereo-s`	Stereo matching	689	284	2.4×	3293	1113	commercial BSD-3-Clause / BSD-3-Clause
`igev-stereo-eth3d`	Stereo matching	764	288	2.6×	1237	1113	commercial MIT / MIT (assumed from code)
`igev-stereo-kitti12`	Stereo matching	764	288	2.6×	1237	1113	commercial MIT / MIT (assumed from code)
`igev-stereo-kitti15`	Stereo matching	764	288	2.6×	1237	1113	commercial MIT / MIT (assumed from code)
`igev-stereo-middlebury`	Stereo matching	764	288	2.6×	1237	1113	commercial MIT / MIT (assumed from code)
`igev-stereo-sceneflow`	Stereo matching	764	288	2.6×	1237	1113	commercial MIT / MIT (assumed from code)
`bridgedepth-sf`	Stereo matching	437	348	1.3×	2916	2310	non-commercial MIT / CC-BY-NC-4.0
`selective-igev-eth3d-finetune`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-eth3d-train`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-kitti12`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-kitti15`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-middlebury-finetune`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-middlebury-train`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`selective-igev-sceneflow`	Stereo matching	998	349	2.9×	1238	1109	commercial MIT / MIT (assumed from code)
`bridgedepth-rvc-pretrain`	Stereo matching	442	354	1.2×	2968	2361	non-commercial MIT / CC-BY-NC-4.0
`cas-gwcnet-gc-sceneflow`	Stereo matching	472	358	1.3×	2338	2272	commercial MIT / MIT (assumed from code)
`s2m2-l`	Stereo matching	1015	376	2.7×	3245	1249	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`marigold-depth-v1-1`	Monocular depth	408	383	1.1×	2817	2583	commercial Apache-2.0 / LicenseRef-CreativeML-OpenRAIL-PlusPlus-M
`waft-stereo-b`	Stereo matching	1137	393	2.9×	5775	1907	non-commercial BSD-3-Clause / CC-BY-NC-4.0
`igev-plusplus-eth3d`	Stereo matching	1081	393	2.7×	1594	1389	commercial MIT / MIT (assumed from code)
`igev-plusplus-kitti12`	Stereo matching	1081	393	2.7×	1594	1389	commercial MIT / MIT (assumed from code)
`igev-plusplus-kitti15`	Stereo matching	1081	393	2.7×	1594	1389	commercial MIT / MIT (assumed from code)
`igev-plusplus-middlebury`	Stereo matching	1081	393	2.7×	1594	1389	commercial MIT / MIT (assumed from code)
`igev-plusplus-sceneflow`	Stereo matching	1081	393	2.7×	1594	1389	commercial MIT / MIT (assumed from code)
`great-igev-eth3d`	Stereo matching	1085	393	2.8×	1402	1192	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`great-igev-kitti`	Stereo matching	1085	393	2.8×	1402	1192	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`great-igev-sceneflow`	Stereo matching	1085	393	2.8×	1402	1192	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`cfnet-sceneflow`	Stereo matching	484	401	1.2×	3441	3662	commercial MIT / MIT (assumed from code)
`mocha-v2-sceneflow`	Stereo matching	1034	448	2.3×	1632	1587	commercial MIT / MIT (assumed from code)
`mocha-v1-sceneflow`	Stereo matching	1057	456	2.3×	1632	1588	commercial MIT / MIT (assumed from code)
`vggt-1b`	Multi-view depth	651	458	1.4×	3691	3702	non-commercial LicenseRef-VGGT / CC-BY-NC-4.0
`vggt-omega-1b-512`	Multi-view depth	726	464	1.6×	3421	3422	non-commercial LicenseRef-FAIR-Noncommercial-Research / LicenseRef-FAIR-Noncommercial-Research
`mapanything-apache`	Multi-view depth	636	544	1.2×	3697	3697	commercial Apache-2.0 / Apache-2.0
`great-igev-middlebury`	Stereo matching	1785	579	3.1×	1875	1553	commercial Apache-2.0 / Apache-2.0 (assumed from code)
`defom-stereo-vitl-sceneflow`	Stereo matching	1531	659	2.3×	4362	3154	non-commercial MIT / CC-BY-NC-4.0
`monster-mix-all`	Stereo matching	1585	668	2.4×	–	–	non-commercial LicenseRef-noncommercial / LicenseRef-noncommercial AND CC-BY-NC-4.0
`monster-sceneflow`	Stereo matching	1585	668	2.4×	2336	2356	non-commercial LicenseRef-noncommercial / LicenseRef-noncommercial AND CC-BY-NC-4.0
`s2m2-xl`	Stereo matching	1815	682	2.7×	5228	2092	non-commercial CC-BY-NC-4.0 / CC-BY-NC-4.0
`monster-plusplus-mix-all`	Stereo matching	1633	698	2.3×	2583	2438	non-commercial LicenseRef-noncommercial / LicenseRef-noncommercial AND CC-BY-NC-4.0
`foundationstereo-vit-s`	Stereo matching	2514	1120	2.2×	2493	2169	non-commercial LicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0 / LicenseRef-NVIDIA-Source-Code-License-NC
`c-foundationstereo-s`	Stereo matching	3646	1362	2.7×	6312	3107	commercial Apache-2.0 / LicenseRef-NVIDIA-Open-Model-License
`foundationstereo-vit-l`	Stereo matching	2990	1513	2.0×	3141	2818	non-commercial LicenseRef-NVIDIA-Source-Code-License-NC AND Apache-2.0 / LicenseRef-NVIDIA-Source-Code-License-NC

Rows keep the reference GPU's kernels-time order on every device; click a header to sort by it. Times are milliseconds per call; – means not timed or not available. Memory is the peak allocated by PyTorch. Licence data comes from the model registry and is not legal advice.