Models · Stereo matching

MonSter++

The accuracy model of MonSter++, which combines mono depth and stereo matching; Autel Robotics and Huazhong University of Science and Technology. Non-commercial licence.

stable Fork of Junda24/MonSter-plusplus@9167101 · backends: portable, kernels

Checkpoint monster-plusplus-mix-all

Paper 2025 Added in 0.1.0 Evaluated with 0.2.0

Kernels, RTX 5060 Ti 16 GB 657 ms 994x1482, fp16
vs the faster original 6.6× original fp16: 4335 ms
Peak memory 2,438 MiB original: 6,382 MiB
Accuracy vs original fp32 ±0.00 kernels bad px %, ETH3D two-view train (1.08 original)

Use it

pip install fastforks

Lean install with only what this model needs: pip install "fastforks-core[nc]"

from fastforks import stereo

model = stereo.load("monster-plusplus-mix-all")  # kernels on a supported GPU, else portable PyTorch
result = model(left, right)
result.disparity
result.backend  # what ran: mode, device, dtype, kernels

The weights are non-commercial: fastforks downloads them only with FASTFORKS_NONCOMMERCIAL=1 set.

fastforks bench monster-plusplus-mix-all

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.

BackenddtypeTime per call (ms)vs original fp16Peak memory
allocated / reserved
Load (s)First call (s)
Original fp32fp32 43391.00×6,382 MiB/6,430 MiB––
Original fp16fp16 4335–6,382 MiB/6,430 MiB––
Portable fp32fp32not measured ––––
Portablefp16 16332.7×2,583 MiB/2,918 MiB2.51.63
Kernelsfp16 6576.6×2,438 MiB/2,754 MiB3.10.66
Portable + torch.compile 11263.8×2,328 MiB/2,532 MiB109 (build)–
TensorRT fp16fp16 6976.2×2,642 MiBenginenot comparable416 (build)–

Measured on RTX 5060 Ti 16 GB, torch 2.14.1+cu130, 2026-11-01 to 2026-11-12. 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

DeviceBackenddtypems / callPeak memoryOutput vs referenceInput
Apple M1 Pro 16 GB MPS, torch 2.14.1, 2026-11-12Originalfp3217464 12.6 GiB processmatches994x1482
Portablefp1611876 7,369 MiB processmatches994x1482
Portablefp3214519 10.8 GiB processmatches994x1482
Kernelsfp1610202 7,347 MiB processmatches994x1482
NVIDIA L4 CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp161742 2,557 MiB/2,832 MiB–994x1482
Kernelsfp16949 2,408 MiB/2,654 MiB–994x1482
NVIDIA L40S CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp16490 2,557 MiB/2,834 MiB–994x1482
Kernelsfp16288 2,408 MiB/2,654 MiB–994x1482
NVIDIA A10 CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp161533 2,557 MiB/2,834 MiB–994x1482
Kernelsfp16792 2,355 MiB/2,714 MiB–994x1482
NVIDIA A100-SXM4-40GB CUDA, torch 2.14.1+cu130, 2026-11-07Portablefp16598 2,557 MiB/2,834 MiB–994x1482
Kernelsfp16289 2,408 MiB/2,654 MiB–994x1482
NVIDIA RTX PRO 6000 CUDA, torch 2.14.1+cu130, 2026-11-08Portablefp16285 2,580 MiB/2,734 MiB–994x1482
Kernelsfp16167 2,432 MiB/2,794 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

Non-commercial only

Code
LicenseRef-noncommercial LICENSE
Weights
LicenseRef-noncommercial AND CC-BY-NC-4.0 source
Training data
mix_all: 'all the datasets we collect over 2M image pairs' (README), not listed
  • code: LicenseRef-noncommercial (noncommercial)
  • weights: LicenseRef-noncommercial AND CC-BY-NC-4.0

Our reading of the licences, not legal advice.

Weights

Mix_all_large.pth 1.6 GB· hosted on Hugging Face

fastforks downloads and checks these files on first use (fastforks download monster-plusplus-mix-all). Non-commercial weights need FASTFORKS_NONCOMMERCIAL=1.

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 % / EPEtraining data not stated: possibly seen

Rundtypebad px % ↓EPE ↓Δ bad px % vs fp32 originalpairsms / pair
Original fp32fp321.080.175 reference 27 818
Original fp16fp161.080.175 ±0.00 27 816
Portable fp32fp321.080.175 ±0.00 27 782
Portablefp161.080.175 ±0.00 27 362
Kernelsfp161.080.175 ±0.00 27 159

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 % / EPEtraining data not stated: possibly seen

Rundtypebad px % ↓EPE ↓Δ bad px % vs fp32 originalpairsms / pair
Original fp32fp323.770.630 reference 15 3556
Original fp16fp163.780.630 −0.01 15 3570
Portable fp32fp323.780.631 −0.01 15 3324
Portablefp163.750.630 +0.02 15 1359
Kernelsfp163.750.630 +0.02 15 556

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| <= 0.2 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.03 px and p99 <= 0.25 px.

CheckToleranceMeasured
CPU fp32, max |diff|≤ 0.2 pxnot recorded
GPU default dtypes, mean |diff|≤ 0.03 px0.02547 px
GPU default dtypes, p99 |diff|≤ 0.25 px0.13171 px

Recorded for monster-plusplus-mix-all on RTX 5060 Ti, torch 2.14.1+cu130, 2026-11-12, with these kernels: cudnn_hacccutlass_convdpt_fusedfastbbigevigevpp

Details

MonSter++ (Cheng et al. 2025), accuracy model (the MonSter network, max disparity 416): Depth Anything V2 (ViT-L) monocular branch + IGEV-style geometry volume, with stereo and mono ConvGRUs that refine each other, 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 (7,929 MiB allocated)
Options
  • iters=32>= 1· changes the output
Outputs
disparity
CUDA kernels
  • igevIGEV stereo: convolution epilogues, correlation lookups, GRU gates
  • igevppIGEV++ and GREAT-IGEV: correlation volume, lookups, attention
  • cutlass_convCUTLASS convolutions with the recurrent update (GRU) fused in
  • cudnn_haccconvolutions with fp16 accumulation (cuDNN)
  • fastbbfast backbone kernels: flash attention, window attention, instance norm and rotary embeddings for ViT and similar backbones
  • dpt_fusedfused fp16 matrix multiplies and convolutions for DINOv2 backbones and DPT depth heads
Apple (MPS)
stereo_metalMetal: stereo kernels. fp32, fp16
Full reference, generated from the repository

The accuracy model of MonSter++, which combines mono depth and stereo matching; Autel Robotics and Huazhong University of Science and Technology. Non-commercial licence.

Commercial use: no (code: LicenseRef-noncommercial (noncommercial); weights: LicenseRef-noncommercial AND CC-BY-NC-4.0).

variantinputbackendtime (ms)peak memory (MiB, allocated / reserved)
monster-plusplus-mix-allMiddlebury 2014 half (994x1482)original (fp32)
original
portable
kernels
portable + torch.compile
TensorRT fp16
4339
4334
1633
657
1126
697
6382 / 6430
6382 / 6430
2583 / 2918
2438 / 2754
2328 / 2532
3489 / 3552
Details
  • Original code: https://github.com/Junda24/MonSter-plusplus at 9167101af8fb
  • Code licence: LicenseRef-noncommercial (noncommercial); files: LICENSE
  • Weights licence: LicenseRef-noncommercial AND CC-BY-NC-4.0
  • Licence evidence: https://github.com/Junda24/MonSter-plusplus/blob/9167101af8fbe93c706707daa7a8181089c8b1b4/LICENSE#L34
  • Training data: mix_all: 'all the datasets we collect over 2M image pairs' (README), not listed
  • Options: iters = 32 (>= 1)
  • Kernel libraries: igev, igevpp, cutlass_conv, cudnn_hacc, fastbb, dpt_fused
  • Variants: monster-plusplus-mix-all
  • 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): monster-plusplus-mix-all 1988x2880 (7.7 GiB peak allocated, 9.6 reserved)
  • monster-plusplus-mix-all 994x1482, first call: portable + torch.compile 109 s (cold Inductor cache), TensorRT fp16 416 s (ONNX + engine).
  • monster-plusplus-mix-all load / first call: original 8.37 s / 1.233 s (375x1242), portable 2.45 s / 1.633† s, kernels 3.12 s / 0.421 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.

variantdatasetoriginal fp32original fp16portableportable fp32kernelsnote
monster-plusplus-mix-allETH3D two-view train, bad > 1 px % / EPE1.08 / 0.1751.08 / 0.1751.08 / 0.1751.08 / 0.1751.08 / 0.175possibly in the training data
monster-plusplus-mix-allMiddlebury 2014 trainingH, bad > 2 px % / EPE3.77 / 0.6303.78 / 0.6303.75 / 0.6303.78 / 0.6313.75 / 0.630possibly in the training data

Measured on NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 468fba10, 2026-11-05.

Parity

CPU, float32: max |diff| <= 0.2 vs the original. GPU, float16 (the default), portable and kernels: mean |diff| <= 0.03 px and p99 <= 0.25 px.

variantkernels error vs original, mean / p99GPU, torchkernels used
monster-plusplus-mix-all0.02547 / 0.13171RTX 5060 Ti, torch 2.14.1+cu130cudnn_hacc@cudnn92400, cutlass_conv@988138e042cb:sm_120, dpt_fused@48e0065bb7cf:sm_120, fastbb@7d058322634d: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).

Left view
Right view
Output
Motorcycle, Middlebury 2014 stereo (left and right view, downscaled 4x), left view
Left view · 100%

Input: Motorcycle, Middlebury 2014 stereo (left and right view, downscaled 4x), Scharstein et al., Middlebury College (research dataset, no licence stated).