RoMa v2 (fastforks)
RoMa v2 (Edstedt, Nordström, Wadenbäck, Felsberg et al. 2025): dense feature matching between two images of any viewpoint. A frozen DINOv3 ViT-L/16 describes both images at a coarse square size, a multi-view ViT with alternating attention and a DPT head turn the global cosine matches into a coarse warp and an overlap logit, and convolutional refiners on VGG19-BN features (local correlation at strides 4, 2, 1) sharpen it, at the coarse size and again at a fine size. Returns, for every pixel of the first image, its position in the second (pixels) and the probability that it is visible there.
This repo is the model card only: it holds no weights. fastforks downloads the authors' weights from the links below and checks them against the SHA-256 in the model card. Code: https://github.com/kleinicke/fastforks. Kernels for the kernels backend: https://huggingface.co/fastforks/kernels. Original code: https://github.com/Parskatt/romav2 at 95c9968145c8.
Task: matching. Tier: experimental. Outputs: warp, certainty. Backends: portable, kernels.
Input: two RGB images of one scene. Output: for every pixel of the first image its position in the second (x, y in pixels) and a certainty that the pixel is visible in both. Sparse matchers (LightGlue) return the keypoints of both images and the index pairs of the matched ones instead. To reconstruct camera poses from an image folder with these matchers, see structure from motion.
Variants
Weights are the authors' files (not hosted here). SHA-256 shortened to 12 hex digits; the full hashes are in card.toml.
| variant | weights (author link) | SHA-256 | weights licence | commercial use |
|---|---|---|---|---|
romav2@1 |
romav2.0.1.pt | 1557dec0d21b |
MIT AND LicenseRef-DINOv3-License | yes |
Licence
Original-code strings, recorded as they are (research-use package; check them before any other use). Code: MIT AND LicenseRef-DINOv3-License. Licence files: LICENSE, LICENSE.md.
Training data: Upstream README: a mix of wide-baseline matching datasets following UFM and MapAnything (MegaDepth and others; see the paper); none has a dataset entry here yet. The frozen DINOv3 descriptor was pretrained on LVD-1689M (not released)..
- Licence note: The checkpoint includes the DINOv3 ViT-L/16 weights, and net/dinov3.py and net/vit/ come from DINOv3: the DINOv3 License (Meta: attribution and conditions apply) covers those parts; read it before commercial use.
Licence evidence: https://github.com/Parskatt/romav2/blob/95c9968145c8906b7b59383258e9f73b02853d89/README.md#license
Input size
any H x W >= 32, padded to a multiple of 32
Tested up to (portable, fp16, RTX 5060 Ti 16 GB): romav2 3000x4000 (2.9 GiB peak allocated, 4.7 reserved)
Speed and memory
Matching, synthetic, 768x1024
| model | time (ms): original fp32 | original | portable | kernels | kernels speed-up: vs original | vs portable | peak memory (MiB, allocated / reserved): original | portable | kernels |
|---|---|---|---|---|---|---|---|---|---|
romav2 |
1412 | 741 | 546 | 354 | 2.1x | 1.5x | 9881 / 10124 | 1989 / 2596 | 1909 / 2300 |
Startup and first call
| model | load (s): original | portable | kernels | first call (s): original | portable | kernels |
|---|---|---|---|---|---|---|
romav2 |
5.45 | 0.85 | 0.96 | 2.09 | 0.91 | 0.62 |
torch.compile and TensorRT
| model | input | time: portable | compile | TensorRT fp16 | kernels | preparation: compile (cold first call) | ONNX + engine build | peak MiB (alloc / reserved): portable | compile | kernels | TensorRT (engine + torch) |
|---|---|---|---|---|---|---|---|---|---|---|---|
romav2 |
768x1024 | 546 ms | 576 ms | 353 ms | 354 ms | 254 s | 185 s | 1989 / 2596 | 3150 / 3386 | 1909 / 2300 | 6694 / 6740 |
Fixed-size cells: time (ms) / peak allocated / peak reserved memory (MiB).
Hardware, protocol and caveats: benchmarks (RTX 5060 Ti 16 GB, sm_120). Original = the original code named above at the precision of portable (fp16, or (fp32) where fp16 is unsupported or not finite); original fp32 = its default fp32 run. Peak memory is the torch allocator's.
Accuracy
On public datasets
Dense matchers: 5000 matches per pair sampled from the warp by certainty (upstream's balanced KDE sampling); sparse matchers: all their matches of one call (LightGlue, XFeat: LightGlue's MegaDepth-1500 setting; LoFTR, DeDoDe: their card defaults). OpenCV essential-matrix RANSAC (0.5 px threshold, normalised by the mean focal length); AUC of the larger of the rotation and translation-direction errors. The same 300 pairs for all modes, so float32 and bfloat16 compare pair for pair. Original = the original code named under Details, same checkpoint and protocol.
| variant | dataset | original fp32 | original fp16 | portable | kernels | note |
|---|---|---|---|---|---|---|
romav2 |
MegaDepth-1500, relative pose AUC@5 / @10 / @20 degrees | 63.2 / 77.1 / 86.7 | 63.0 / 76.9 / 86.5 | 62.9 / 76.9 / 86.4 | ||
romav2 |
MegaDepth-1500, every 5th pair, relative pose AUC@5 / @10 / @20 degrees | 64.6 / 78.0 / 87.1 | 64.7 / 78.2 / 87.3 | 64.7 / 78.3 / 87.3 | 64.4 / 77.9 / 86.9 |
Measured on portable: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 011e320c, 2026-11-11 to 2026-11-12; kernels: NVIDIA GeForce RTX 5060 Ti, torch 2.14.1+cu130, fastforks 011e320c, 2026-11-12.
Parity with the original code
Every variant is checked against reference outputs of the original code at its pinned commit. CPU, float32: max |diff| <= 1.6e-05 vs the original. GPU, bfloat16 (the default), portable and kernels: mean |diff| <= 0.0086 and p99 <= 0.032 (absolute error of the output warp).
Limitations
- The precision matrices of the warp (upstream
precision_AB) and upstream's sampling / keypoint helpers are not returned; tools/eval_matching.py has the MegaDepth-1500 sampling. - Both images are resized to squares (upstream's behaviour), whatever their aspect ratio.
Usage
from fastforks import matching
model = matching.load("romav2") # mode="auto": the preferred backend from recorded measurements
result = model("view1.jpg", "view2.jpg")
result.warp # [2, H1, W1] float32: x, y in view2 pixels
result.backend # what ran, e.g. Backend(mode='portable', device='cuda:0', dtype='float16')
Install with pip install fastforks (see https://github.com/kleinicke/fastforks). Other variants: romav2.
Kernels backend
matching.load("romav2", mode="kernels") runs our CUDA kernels (romav2). It needs a prebuilt kernel for your GPU architecture (downloaded from https://huggingface.co/fastforks/kernels on first use) and a recent NVIDIA driver; if there is none it raises BackendUnavailableError with the reason and never falls back silently. mode="auto" picks kernels when it can run and logs the choice.