GREAT-IGEV
IGEV-Stereo extended with global attention (spatial, epipolar-row and volume attention).
stable Fork of JarvisLee0423/GREAT-Stereo@2a6b1ef · backends: portable, kernels
- Paper Global Regulation and Excitation via Attention Tuning for Stereo Matching ICCV 2025 · first released 19 Sept 2025
Paper 2025 Added in 0.1.0 Evaluated with 0.2.0
Use it
pip install fastforks Lean install with only what this model needs: pip install fastforks-core
from fastforks import stereo
model = stereo.load("great-igev-sceneflow") # kernels on a supported GPU, else portable PyTorch
result = model(left, right)
result.disparity
result.backend # what ran: mode, device, dtype, kernels fastforks bench great-igev-sceneflow Times the model on this machine. load(…, mode="portable") or mode="kernels" picks a backend; usage guide.
Speed and memory
Every backend on one GPU (RTX 5060 Ti 16 GB) and the same input, the authors’ code included. Kernels runs fp16. Portable runs fp16 on CUDA.
| Backend | dtype | Time per call (ms) | vs original fp16 | Peak memory allocated / reserved | Load (s) | First call (s) |
|---|---|---|---|---|---|---|
| Original fp32measured on great-igev-eth3d (same architecture, other checkpoint) | fp32 | 2252 | 0.65× | 8,964 MiB/9,268 MiB | – | – |
| Original fp16measured on great-igev-eth3d (same architecture, other checkpoint) | fp16 | 1458 | – | 8,861 MiB/9,206 MiB | – | – |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portablemeasured on great-igev-eth3d (same architecture, other checkpoint) | fp16 | 1085 | 1.3× | 1,402 MiB/1,678 MiB | 0.3 | 1.10 |
| Kernels | fp16 | 397 | 3.7× | 1,134 MiB/1,302 MiB | 0.6 | 0.40 |
| Portable + torch.compile | 830 | 1.8× | 2,831 MiB/3,362 MiB | 224 (build) | – | |
| TensorRT fp16 | fp16 | 944 | 1.5× | 6,020 MiBenginenot comparable | 675 (build) | – |
Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-03 to 2026-11-16. First call, 3 warm-up, peak reset, median of 3 rounds x median of 5. Ratios divide two
times of this table (same GPU, same input). Original fp16 is the authors’ own precision option, else torch.autocast. torch.compile and TensorRT: build time instead of load time. TensorRT memory (engine, not comparable) is the GPU memory the engine took, read from the driver because TensorRT allocates outside PyTorch, so it can't be compared with the PyTorch peaks of the other rows.
Other devices
| Device | Backend | dtype | ms / call | Peak memory | Output vs reference | Input |
|---|---|---|---|---|---|---|
| Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-13 | Original | fp32 | out of memory | – | 994x1482 | |
| Portable | fp16 | 11062 | 9,960 MiB process | matches | 994x1482 | |
| Portable | fp32 | 32187 | 12.4 GiB process | matches | 994x1482 | |
| Kernels | fp16 | 9861 | 10,126 MiB process | matches | 994x1482 | |
| NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 1235 | 1,375 MiB/1,558 MiB | – | 994x1482 |
| Kernels | fp16 | 579 | 1,195 MiB/1,602 MiB | – | 994x1482 | |
| NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 354 | 1,375 MiB/1,558 MiB | – | 994x1482 |
| Kernels | fp16 | 176 | 1,195 MiB/1,582 MiB | – | 994x1482 | |
| NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 982 | 1,375 MiB/1,558 MiB | – | 994x1482 |
| Kernels | fp16 | 477 | 1,195 MiB/1,622 MiB | – | 994x1482 | |
| NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 383 | 1,375 MiB/1,558 MiB | – | 994x1482 |
| Kernels | fp16 | 196 | 1,195 MiB/1,582 MiB | – | 994x1482 | |
| NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 191 | 1,399 MiB/1,598 MiB | – | 994x1482 |
| Kernels | fp16 | 98.3 | 1,189 MiB/1,602 MiB | – | 994x1482 | |
CUDA memory is torch’s peak allocated / reserved; Apple has no such counter, so the Mac rows show the process footprint (weights, Python and torch included). “Output vs reference” compares the device’s output with the stored golden output under the card’s tolerance; – = not checked on that device. Kernels “not implemented” = no kernels for this model on that backend yet: it runs portable there.
Licence and weights
Commercial use allowed
- Code
- Apache-2.0 LICENSE
- Weights
- Apache-2.0 (assumed from code)(not stated by the authors; assumed to follow the code) source
- Training data
-
- Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa) research-only, non-commercial
- weights licence not stated by the original authors; assumed to be the code licence
Our reading of the licences, not legal advice.
Weights
great-igev-sceneflow-submit.pth 58.4 MB· hosted on Google Drivefastforks downloads and checks these files on first use (fastforks download great-igev-sceneflow).
Accuracy
The test set through the authors’ code and through ours: same checkpoint, same protocol. The distance between the original’s fp32 and fp16 runs is what half precision alone costs; our runs should stay within it.
ETH3D two-view train
bad > 1 px % / EPE
| Run | dtype | bad px % ↓ | EPE ↓ | Δ bad px % vs fp32 original | pairs | ms / pair |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 4.48 | 0.364 | reference | 27 | 545 |
| Original fp16 | fp16 | 4.41 | 0.363 | +0.07 | 27 | 367 |
| Portable fp32 | fp32 | 4.45 | 0.364 | +0.03 | 27 | 482 |
| Portable | fp16 | 4.43 | 0.364 | +0.05 | 27 | 253 |
| Kernels | fp16 | 4.44 | 0.364 | +0.04 | 27 | 109 |
Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-05; portable fp32, portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05.
Middlebury 2014 trainingH
bad > 2 px % / EPE
| Run | dtype | bad px % ↓ | EPE ↓ | Δ bad px % vs fp32 original | pairs | ms / pair |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 11.22 | 2.710 | reference | 15 | 1913 |
| Original fp16 | fp16 | 11.19 | 2.698 | +0.03 | 15 | 1254 |
| Portable fp32 | fp32 | 11.20 | 2.706 | +0.02 | 15 | 1825 |
| Portable | fp16 | 11.14 | 2.687 | +0.08 | 15 | 923 |
| Kernels | fp16 | 11.16 | 2.691 | +0.06 | 15 | 338 |
Original = the authors’ code, same checkpoint and protocol. Measured on original fp32, original fp16: RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-05; portable fp32, portable, kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05.
Closeness to the original
Beyond the test-set scores: the raw output of ours against the original’s fp32 output on the same input, pixel by pixel. 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.
| Check | Tolerance | Measured |
|---|---|---|
| CPU fp32, max |diff| | ≤ 0.002 px | not recorded |
| GPU default dtypes, mean |diff| | ≤ 0.02 px | 0.00166 px |
| GPU default dtypes, p99 |diff| | ≤ 0.25 px | 0.02857 px |
Recorded for great-igev-sceneflow on RTX 5060 Ti, torch 2.14.1+cu130, 2026-10-31, with these kernels: cudnn_hacccutlass_convigevigevpp
Details
GREAT-IGEV (Li et al. 2025): IGEV-Stereo with global attention (spatial, epipolar-row matching and volume attention) that builds an excitation-weighted cost volume, then IGEV's ConvGRU updates, 32 iterations.
- Input / output
- Input: a rectified left / right image pair. Output: a disparity map (pixels, left view).
- Input size
- any H x W >= 32, padded to a multiple of 32; tested up to 3000x4000 (11.1 GiB allocated)
- Options
iters=32>= 1· changes the outputcuda_graphs=false
- Outputs
disparity- CUDA kernels
igevIGEV stereo: convolution epilogues, correlation lookups, GRU gatesigevppIGEV++ and GREAT-IGEV: correlation volume, lookups, attentioncutlass_convCUTLASS convolutions with the recurrent update (GRU) fused incudnn_haccconvolutions with fp16 accumulation (cuDNN)
- Apple (MPS)
stereo_metalMetal: stereo kernels. fp32, fp16
Full reference, generated from the repository
IGEV-Stereo extended with global attention (spatial, epipolar-row and volume attention).
Commercial use: yes.
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
great-igev-sceneflow | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2252 1458 1085 397 830 944 | 8964 / 9268 8861 / 9206 1402 / 1678 1134 / 1302 2831 / 3362 6155 / 6234 |
great-igev-kitti | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2252 1458 1085 396 830 942 | 8964 / 9268 8861 / 9206 1402 / 1678 1134 / 1302 2831 / 3362 6155 / 6234 |
great-igev-eth3d | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2252 1458 1085 397 829 940 | 8964 / 9268 8861 / 9206 1402 / 1678 1134 / 1302 2831 / 3362 6151 / 6172 |
great-igev-middlebury (max_disp 768) | Middlebury 2014 half (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | OOM OOM 1785 580 1676 fails: AttributeError: 'NoneType' object has no attribute 'get_tens | – – 1875 / 2078 1472 / 1642 5228 / 5720 – |
Details
- Original code: https://github.com/JarvisLee0423/GREAT-Stereo at
2a6b1efede8c - Code licence: Apache-2.0 (permissive); files:
LICENSE - Weights licence: Apache-2.0 (assumed from code)
- Licence evidence: https://github.com/JarvisLee0423/GREAT-Stereo/blob/2a6b1efede8c3a9eec9540790c2700ea2f3dee7d/README.md#L558
- Training data: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa)
great-igev-kitti: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), KITTI stereo / optical flow 2012 / 2015 (SceneFlow, then fine-tuned on KITTI 2012 + 2015 (launchers/.../great_igev_trainer_kitti.sh; KITTI submission model))great-igev-eth3d: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), CREStereo synthetic stereo dataset (Megvii), InStereo2K, ETH3D two-view stereo, not stated by the authors (ETH3D submission model; original recipe (launchers, utils/stereo_matching/data_utils/data_fetchers.py): SceneFlow model fine-tuned on eth3d_finetune (ETH3D, InStereo2K, CREStereo); whether the eth3d_train mixture (+ TartanAir, Sintel) ran first is not stated)great-igev-middlebury: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), CREStereo synthetic stereo dataset (Megvii), InStereo2K, HR-VS (high-resolution virtual stereo, Yang et al. CVPR 2019), Middlebury stereo (2014), NVIDIA Falling Things (FAT), not stated by the authors (Middlebury submission model; original recipe (launchers, utils/stereo_matching/data_utils/data_fetchers.py): SceneFlow model fine-tuned on middlebury_finetune (CREStereo, InStereo2K, HR-VS, Middlebury 2005 / 2006 / 2014 / 2021 / MiddEval3 H + F, Falling Things); whether the middlebury_train mixture (+ TartanAir) ran first is not stated)
- Options:
iters= 32 (>= 1),cuda_graphs= False (speed only) - Kernel libraries:
igev,igevpp,cutlass_conv,cudnn_hacc - Variants:
great-igev-sceneflow,great-igev-kitti,great-igev-eth3d,great-igev-middlebury - Outputs: disparity
- Input size: any H x W >= 32, padded to a multiple of 32
- Tested up to (portable, fp16, RTX 5060 Ti 16 GB):
great-igev-sceneflow3000x4000 (11.1 GiB peak allocated, 14.8 reserved);great-igev-kitti3000x4000 (11.1 GiB peak allocated, 14.8 reserved);great-igev-eth3d3000x4000 (11.1 GiB peak allocated, 14.8 reserved);great-igev-middlebury1988x2880 (7.1 GiB peak allocated, 11.3 reserved) great-igev-sceneflow994x1482, first call: portable + torch.compile 224 s (cold Inductor cache), TensorRT fp16 675 s (ONNX + engine).great-igev-kitti994x1482, first call: portable + torch.compile 225 s (cold Inductor cache), TensorRT fp16 687 s (ONNX + engine).great-igev-eth3d994x1482, first call: portable + torch.compile 224 s (cold Inductor cache), TensorRT fp16 685 s (ONNX + engine).great-igev-middlebury994x1482, first call: portable + torch.compile 326 s (cold Inductor cache).great-igev-scenefloworiginal (fp32) 994x1482, original 994x1482, portable 994x1482: measured on great-igev-eth3d (same architecture, other checkpoint)great-igev-kittioriginal (fp32) 994x1482, original 994x1482, portable 994x1482: measured on great-igev-eth3d (same architecture, other checkpoint)great-igev-sceneflowload / first call: original 0.44 s / 4.596 s (375x1242), portable 0.32 s / 1.102† s, kernels 0.64 s / 0.702 s (375x1242)great-igev-kittiload / first call: original 0.44 s / 4.596 s (375x1242), portable 0.32 s / 1.102† s, kernels 0.65 s / 0.699 s (375x1242)great-igev-eth3dload / first call: original 0.44 s / 4.596 s (375x1242), portable 0.32 s / 1.102† s, kernels 0.7 s / 1.138 s (375x1242)great-igev-middleburyload / first call: original 0.39 s / 4.586 s (375x1242), portable 3.91 s / 1.808† s, kernels 1.02 s / 0.732 s (375x1242)- † measured in the speed sweep after the portable load (or an earlier size) in the same process: file cache and kernels are warm, so the time is lower than a fresh start.
- Protocol, median of 5 after 3 warm-up, uint8 CUDA input (synthetic for bench_inputs sizes), card defaults; gc before each load; peak = max over the first call and the timed calls
Measured
Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | portable fp32 | kernels | note |
|---|---|---|---|---|---|---|---|
great-igev-sceneflow | ETH3D two-view train, bad > 1 px % / EPE | 4.48 / 0.364 | 4.41 / 0.363 | 4.43 / 0.364 | 4.45 / 0.364 | 4.44 / 0.364 | |
great-igev-sceneflow | Middlebury 2014 trainingH, bad > 2 px % / EPE | 11.22 / 2.710 | 11.19 / 2.698 | 11.14 / 2.687 | 11.20 / 2.706 | 11.16 / 2.691 | |
great-igev-kitti | ETH3D two-view train, bad > 1 px % / EPE | 23.35 / 3.115 | 23.36 / 3.116 | 23.37 / 3.113 | 23.37 / 3.120 | 23.36 / 3.121 | |
great-igev-kitti | Middlebury 2014 trainingH, bad > 2 px % / EPE | 17.99 / 4.677 | 18.00 / 4.679 | 17.98 / 4.670 | 17.99 / 4.679 | 18.00 / 4.676 | |
great-igev-eth3d | ETH3D two-view train, bad > 1 px % / EPE | 0.49 / 0.116 | 0.49 / 0.117 | 0.49 / 0.118 | 0.50 / 0.116 | 0.49 / 0.118 | trained on this dataset's public images; not a zero-shot result |
great-igev-eth3d | Middlebury 2014 trainingH, bad > 2 px % / EPE | 11.20 / 2.894 | 11.21 / 2.903 | 11.19 / 2.946 | 11.19 / 2.923 | 11.21 / 2.883 | possibly in the training data |
great-igev-middlebury | ETH3D two-view train, bad > 1 px % / EPE | 27.42 / 1.133 | 27.43 / 1.134 | 27.44 / 1.131 | 27.40 / 1.132 | 27.53 / 1.135 | possibly in the training data |
great-igev-middlebury | Middlebury 2014 trainingH, bad > 2 px % / EPE | out of memory | out of memory | 4.14 / 0.707 | 4.11 / 0.707 | 4.14 / 0.708 | trained on this dataset's public images; not a zero-shot result |
Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05.
Parity
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.
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
great-igev-sceneflow | 0.00166 / 0.02857 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
great-igev-kitti | 0.00284 / 0.03707 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
great-igev-eth3d | 0.00214 / 0.03202 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
great-igev-middlebury | 0.00162 / 0.00911 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
Example output
One fixed input through this checkpoint, the same for every model of the task (kernels · float16 · NVIDIA GeForce RTX 5060 Ti).
Input: Motorcycle, Middlebury 2014 stereo (left and right view, downscaled 4x), Scharstein et al., Middlebury College (research dataset, no licence stated).