IGEV++
The successor of IGEV-Stereo by Gangwei Xu et al., with multi-range geometry encoding volumes for large disparity ranges.
stable Fork of gangweix/IGEV-plusplus@099dec9 · backends: portable, kernels
- Paper IGEV++: Iterative Multi-range Geometry Encoding Volumes for Stereo Matching IEEE TPAMI 2025 · first released 1 Sept 2024
Paper 2024 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("igev-plusplus-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 igev-plusplus-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 igev-plusplus-eth3d (same architecture, other checkpoint) | fp32 | 2012 | 0.64× | 4,152 MiB/4,242 MiB | – | – |
| Original fp16measured on igev-plusplus-eth3d (same architecture, other checkpoint) | fp16 | 1278 | – | 2,891 MiB/3,108 MiB | – | – |
| Portable fp32 | fp32 | not measured | – | – | – | – |
| Portablemeasured on igev-plusplus-eth3d (same architecture, other checkpoint) | fp16 | 1079 | 1.2× | 1,583 MiB/1,696 MiB | 0.9 | 1.08 |
| Kernelsmeasured on igev-plusplus-eth3d (same architecture, other checkpoint) | fp16 | 395 | 3.2× | 1,272 MiB/1,514 MiB | 0.6 | 0.39 |
| Portable + torch.compile | 718 | 1.8× | 917 MiB/1,054 MiB | 158 (build) | – | |
| TensorRT fp16 | fp16 | 495 | 2.6× | 3,686 MiBenginenot comparable | 426 (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, 2026-11-12 | Original | fp32 | 10992 | 7,860 MiB process | matches | 994x1482 |
| Portable | fp16 | 8602 | 6,037 MiB process | matches | 994x1482 | |
| Portable | fp32 | 10424 | 9,132 MiB process | matches | 994x1482 | |
| Kernels | fp16 | 6135 | 4,972 MiB process | matches | 994x1482 | |
| NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 1202 | 1,568 MiB/1,736 MiB | – | 994x1482 |
| Kernels | fp16 | 547 | 1,333 MiB/1,594 MiB | – | 994x1482 | |
| NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 337 | 1,568 MiB/1,736 MiB | – | 994x1482 |
| Kernels | fp16 | 177 | 1,333 MiB/1,674 MiB | – | 994x1482 | |
| NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 953 | 1,568 MiB/1,736 MiB | – | 994x1482 |
| Kernels | fp16 | 463 | 1,333 MiB/1,756 MiB | – | 994x1482 | |
| NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-07 | Portable | fp16 | 384 | 1,568 MiB/1,736 MiB | – | 994x1482 |
| Kernels | fp16 | 186 | 1,333 MiB/1,674 MiB | – | 994x1482 | |
| NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08 | Portable | fp16 | 185 | 1,591 MiB/1,876 MiB | – | 994x1482 |
| Kernels | fp16 | 104 | 1,327 MiB/1,574 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
- MIT LICENSE
- Weights
- MIT (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
sceneflow.pth 58.5 MB· hosted on Google Drivefastforks downloads and checks these files on first use (fastforks download igev-plusplus-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 | 4.66 | 0.356 | reference | 27 | 482 |
| Original fp16 | fp16 | 4.66 | 0.355 | ±0.00 | 27 | 337 |
| Portable fp32 | fp32 | 4.66 | 0.355 | ±0.00 | 27 | 453 |
| Portable | fp16 | 4.65 | 0.355 | +0.01 | 27 | 250 |
| Kernels | fp16 | 4.66 | 0.356 | ±0.00 | 27 | 123 |
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: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05; kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
Middlebury 2014 trainingH
bad > 2 px % / EPE
| Run | dtype | bad px % ↓ | EPE ↓ | Δ bad px % vs fp32 original | pairs | ms / pair |
|---|---|---|---|---|---|---|
| Original fp32 | fp32 | 10.57 | 3.120 | reference | 15 | 1710 |
| Original fp16 | fp16 | 10.58 | 3.120 | −0.01 | 15 | 1103 |
| Portable fp32 | fp32 | 10.57 | 3.119 | ±0.00 | 15 | 1754 |
| Portable | fp16 | 10.55 | 3.128 | +0.02 | 15 | 917 |
| Kernels | fp16 | 10.56 | 3.153 | +0.01 | 15 | 337 |
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: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05; kernels: RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
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 | 0.00074 px |
| GPU default dtypes, p99 |diff| | ≤ 0.25 px | 0.00835 px |
| real_gpu_mean | ≤ 0.1 px | not recorded |
| real_gpu_p99 | ≤ 2 px | not recorded |
Recorded for igev-plusplus-sceneflow on RTX 5060 Ti, torch 2.14.1+cu130, 2026-10-31, with these kernels: cudnn_hacccutlass_convigevigevpp
Details
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.
- 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 1988x2880 (6,166 MiB allocated)
- Options
iters=32>= 1· changes the outputcuda_graphs=true
- 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
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.
Full reference, generated from the repository
The successor of IGEV-Stereo by Gangwei Xu et al., with multi-range geometry encoding volumes for large disparity ranges.
Commercial use: yes.
| variant | input | backend | time (ms) | peak memory (MiB, allocated / reserved) |
|---|---|---|---|---|
igev-plusplus-sceneflow | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2012 1278 1079 395 718 495 | 4152 / 4242 2891 / 3108 1583 / 1696 1272 / 1514 917 / 1054 3821 / 3900 |
igev-plusplus-kitti12 | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2012 1278 1080 394 718 494 | 4152 / 4242 2891 / 3108 1583 / 1696 1272 / 1514 917 / 1054 3821 / 3900 |
igev-plusplus-kitti15 | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2012 1278 1080 394 718 494 | 4152 / 4242 2891 / 3108 1583 / 1696 1272 / 1494 917 / 1052 3821 / 3900 |
igev-plusplus-middlebury | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2012 1278 1080 393 718 494 | 4152 / 4242 2891 / 3108 1583 / 1696 1272 / 1494 917 / 1054 3821 / 3900 |
igev-plusplus-eth3d | synthetic (994x1482) | original (fp32) original portable kernels portable + torch.compile TensorRT fp16 | 2012 1278 1079 427 716 494 | 4152 / 4242 2891 / 3108 1583 / 1696 1272 / 1514 952 / 1074 3765 / 3786 |
Details
- Original code: https://github.com/gangweix/IGEV-plusplus at
099dec979897 - Code licence: MIT (permissive); files:
LICENSE - Weights licence: MIT (assumed from code)
- Licence evidence: https://github.com/gangweix/IGEV-plusplus/blob/099dec97989780d151700459c682df1cc2c18e88/README.md#L23
- Training data: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa)
igev-plusplus-kitti12: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), KITTI stereo / optical flow 2012 / 2015 (SceneFlow, then fine-tuned on KITTI 2012 + 2015 (README recipe; KITTI 2012 model))igev-plusplus-kitti15: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), KITTI stereo / optical flow 2012 / 2015 (SceneFlow, then fine-tuned on KITTI 2012 + 2015 (README recipe; KITTI 2015 model))igev-plusplus-middlebury: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), TartanAir, NVIDIA Falling Things (FAT), CREStereo synthetic stereo dataset (Megvii), InStereo2K, Middlebury stereo (2014), not stated by the authors (Middlebury submission model; original recipe (README, core/stereo_datasets.py): SceneFlow model, then TartanAir, SceneFlow, Falling Things, InStereo2K, CARLA, CREStereo and Middlebury 2005 / 2006 / 2014 / 2021 / MiddEval3, then a fine-tune on CREStereo, InStereo2K, CARLA, Falling Things and Middlebury)igev-plusplus-eth3d: Freiburg SceneFlow (FlyingThings3D, Driving, Monkaa), TartanAir, MPI Sintel stereo, CREStereo synthetic stereo dataset (Megvii), InStereo2K, ETH3D two-view stereo (ETH3D submission model; original recipe (README, core/stereo_datasets.py): SceneFlow model, then TartanAir, SceneFlow, Sintel, CREStereo, InStereo2K and ETH3D, then a fine-tune on CREStereo, InStereo2K and ETH3D)
- Options:
iters= 32 (>= 1),cuda_graphs= True (speed only) - Kernel libraries:
igev,igevpp,cutlass_conv,cudnn_hacc - Variants:
igev-plusplus-sceneflow,igev-plusplus-kitti12,igev-plusplus-kitti15,igev-plusplus-middlebury,igev-plusplus-eth3d - 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):
igev-plusplus-sceneflow1988x2880 (6.0 GiB peak allocated, 10.2 reserved);igev-plusplus-kitti121988x2880 (6.0 GiB peak allocated, 10.2 reserved);igev-plusplus-kitti151988x2880 (6.0 GiB peak allocated, 10.2 reserved);igev-plusplus-middlebury1988x2880 (6.0 GiB peak allocated, 10.2 reserved);igev-plusplus-eth3d1988x2880 (6.0 GiB peak allocated, 10.2 reserved) igev-plusplus-sceneflow994x1482, first call: portable + torch.compile 158 s (cold Inductor cache), TensorRT fp16 426 s (ONNX + engine).igev-plusplus-kitti12994x1482, first call: portable + torch.compile 158 s (cold Inductor cache), TensorRT fp16 422 s (ONNX + engine).igev-plusplus-kitti15994x1482, first call: portable + torch.compile 159 s (cold Inductor cache), TensorRT fp16 425 s (ONNX + engine).igev-plusplus-middlebury994x1482, first call: portable + torch.compile 157 s (cold Inductor cache), TensorRT fp16 436 s (ONNX + engine).igev-plusplus-eth3d994x1482, first call: portable + torch.compile 159 s (cold Inductor cache), TensorRT fp16 425 s (ONNX + engine).igev-plusplus-scenefloworiginal (fp32) 994x1482, original 994x1482, portable 994x1482, kernels 994x1482: measured on igev-plusplus-eth3d (same architecture, other checkpoint)igev-plusplus-kitti12original (fp32) 994x1482, original 994x1482, portable 994x1482, kernels 994x1482: measured on igev-plusplus-eth3d (same architecture, other checkpoint)igev-plusplus-kitti15original (fp32) 994x1482, original 994x1482, portable 994x1482, kernels 994x1482: measured on igev-plusplus-eth3d (same architecture, other checkpoint)igev-plusplus-middleburyoriginal (fp32) 994x1482, original 994x1482, portable 994x1482, kernels 994x1482: measured on igev-plusplus-eth3d (same architecture, other checkpoint)igev-plusplus-sceneflowload / first call: original 0.42 s / 2.335 s (375x1242), portable 0.9 s / 0.822 s (375x1242), kernels 0.56 s / 0.535 s (375x1242)igev-plusplus-kitti12load / first call: original 0.42 s / 2.335 s (375x1242), portable 0.99 s / 0.819 s (375x1242), kernels 0.59 s / 0.532 s (375x1242)igev-plusplus-kitti15load / first call: original 0.42 s / 2.335 s (375x1242), portable 0.67 s / 0.822 s (375x1242), kernels 0.6 s / 0.534 s (375x1242)igev-plusplus-middleburyload / first call: original 0.42 s / 2.335 s (375x1242), portable 0.69 s / 0.819 s (375x1242), kernels 0.59 s / 0.536 s (375x1242)igev-plusplus-eth3dload / first call: original 0.42 s / 2.335 s (375x1242), portable 0.69 s / 0.839 s (375x1242), kernels 0.68 s / 0.567 s (375x1242)- Protocol, first call, 3 warm-up, peak reset, median of 3 rounds x median of 5
Measured
Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | portable fp32 | 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 / 0.355 | 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 | 10.57 / 3.119 | 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 | 26.74 / 1.967 | 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 / 4.991 | 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 | 25.52 / 1.893 | 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 | 20.70 / 4.848 | 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 | 18.38 / 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 | 4.71 / 0.705 | 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 | 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 / 2.657 | 10.60 / 2.734 |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05; portable32: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 939adfe8, 2026-11-06.
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. On real pairs up to 720x1280 the limits are mean 0.1 / p99 2.0 (precision-sensitive there, see Limitations).
| variant | kernels error vs original, mean / p99 | GPU, torch | kernels used |
|---|---|---|---|
igev-plusplus-sceneflow | 0.00074 / 0.00835 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
igev-plusplus-kitti12 | 0.00116 / 0.00274 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
igev-plusplus-kitti15 | 0.00189 / 0.00365 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
igev-plusplus-middlebury | 0.00064 / 0.00229 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
igev-plusplus-eth3d | 0.00081 / 0.00327 | RTX 5060 Ti, torch 2.14.1+cu130 | cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, igev@f075827d68c6:sm_120, igevpp@df1bb4dbe09a:sm_120 |
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.
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).