romav2 2.0.0__tar.gz

romav2-2.0.0/PKG-INFO

@@ -0,0 +1,161 @@
+Metadata-Version: 2.3
+Name: romav2
+Version: 2.0.0
+Summary: RoMa v2: Harder Better Faster Denser Feature Matching
+Author: Johan Edstedt
+Author-email: Johan Edstedt <johan.edstedt@liu.se>
+Requires-Dist: dataclasses>=0.8
+Requires-Dist: einops>=0.8.1
+Requires-Dist: pillow>=12.0.0
+Requires-Dist: rich>=14.2.0
+Requires-Dist: torch
+Requires-Dist: torchvision>=0.23.0
+Requires-Dist: tqdm>=4.67.1
+Requires-Dist: slurm-util>=0.2.7 ; extra == 'dev'
+Requires-Dist: ruff>=0.14.5 ; extra == 'dev'
+Requires-Dist: kornia>=0.8.2 ; extra == 'eval'
+Requires-Dist: matplotlib>=3.10.7 ; extra == 'eval'
+Requires-Dist: opencv-python>=4.12.0.88 ; extra == 'eval'
+Requires-Dist: wandb>=0.23.0 ; extra == 'eval'
+Requires-Dist: wxbs-benchmark>=0.0.4 ; extra == 'eval'
+Requires-Dist: fused-local-corr ; sys_platform == 'linux' and extra == 'fused-local-corr'
+Requires-Python: >=3.10
+Provides-Extra: dev
+Provides-Extra: eval
+Provides-Extra: fused-local-corr
+Description-Content-Type: text/markdown
+
+<p align="center">
+  <h1 align="center"> <ins>RoMa v2</ins> 🤖: Harder Better Faster Denser Feature Matching
+  <h2 align="center">
+    <a href="https://scholar.google.com/citations?user=Ul-vMR0AAAAJ">Johan Edstedt</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=-vJPE04AAAAJ">David Nordström</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=mvY4rdIAAAAJ">Yushan Zhang</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=FUE3Wd0AAAAJ">Georg Bökman</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=dsEPAvUAAAAJ">Jonathan Astermark</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=vHeD0TYAAAAJ">Viktor Larsson</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=9j-6i_oAAAAJ&hl">Anders Heyden</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=P_w6UgMAAAAJ&hl">Fredrik Kahl</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=6WRQpCQAAAAJ">Mårten Wadenbäck</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=lkWfR08AAAAJ">Michael Felsberg</a>
+  </p>
+  <h2 align="center"><p>
+    <a href="https://arxiv.org/abs/2511.15706" align="center">Paper</a> | 
+    <a href="TBD" align="center">Project Page</a>
+  </p></h2>
+  <div align="center"></div>
+</p>
+<br/>
+<p align="center">
+    <img src="assets/qualitative.png" alt="example" width=80%>
+</p>
+
+## How to Use
+```python
+from romav2 import RoMaV2
+
+# load pretrained model
+model = RoMaV2()
+# Match densely for any image-like pair of inputs
+preds = model.match(img_A_path, img_B_path)
+
+# you can also run the forward method directly as 
+# preds = model(img_A, img_B)
+
+# Sample 5000 matches for estimation
+matches, overlaps, precision_AB, precision_BA = model.sample(preds, 5000)
+
+# Convert to pixel coordinates (RoMaV2 produces matches in [-1,1]x[-1,1])
+kptsA, kptsB = model.to_pixel_coordinates(matches, H_A, W_A, H_B, W_B)
+
+# Find a fundamental matrix (or anything else of interest)
+F, mask = cv2.findFundamentalMat(
+    kptsA.cpu().numpy(), kptsB.cpu().numpy(), ransacReprojThreshold=0.2, method=cv2.USAC_MAGSAC, confidence=0.999999, maxIters=10000
+)
+```
+We additionally provide two demos in the [demos folder](demo), which might help in understanding.
+
+
+## Setup/Install
+In your python environment (tested on Linux python 3.12), run:
+```bash
+uv pip install -e .
+```
+or 
+```bash
+uv sync
+```
+
+## Benchmarks
+If you do not already have MegaDepth and ScanNet, you can the following to download them:
+```bash
+source scripts/eval_prep.sh
+```
+### Mega-1500
+```bash
+uv run tests/test_mega1500.py
+```
+### ScanNet-1500
+```bash
+uv run tests/test_scannet1500.py
+```
+### Expected Results
+Experiments on ScanNet-1500 and MegaDepth-1500 are provided in the [tests folder](tests).
+Running these gave me `ScanNet-1500: [34.0, 56.5, 73.9]`, and `Mega-1500: [62.8, 76,8, 86.5]`, which are similar to the results of the paper.
+
+
+## Fused local correlation kernel
+Include the `--extra fused-local-corr` flag as:
+```bash
+uv sync --extra fused-local-corr
+```
+or 
+```bash
+uv pip install romav2[fused-local-corr]
+```
+or
+```bash
+uv add romav2[fused-local-corr]
+```
+
+## Settings
+By twiddling with some different settings you may reach better results on your task of interest.
+Some important ones, which we enable setting to some reasonable defaults through `model.apply_setting`, are:
+
+`model.H_lr, model.W_lr`: height and width for the image pair.
+
+`model.H_hr, model.W_hr`: height and width for a high resolution version of the image pair (used for upsampling as in RoMa)
+
+`model.bidirectional`: Useful for getting more diverse matches, and for estimating the covariance matrix in both directions.
+
+`model.threshold`: Value between [0,1]. Used to set overlap prediction above it to 1. Useful for Mega1500.
+
+`model.balanced_sampling`: Diverse sampling, same as RoMa. Typically helps to get better RANSAC estimates.
+
+## License
+All our code except DINOv3 is MIT license.
+DINOv3 has a custom license, see [DINOv3](https://github.com/facebookresearch/dinov3/tree/main?tab=License-1-ov-file#readme).
+
+## Acknowledgement
+Our codebase builds mainly on the code in [RoMa](https://github.com/Parskatt/RoMa).
+We were additionally inspired by [UFM](https://github.com/UniFlowMatch/UFM) and [MapAnything](https://github.com/facebookresearch/map-anything), particularly for the datasets used to train the models.
+
+## BibTeX
+If you find our models useful, please consider citing our paper!
+```
+@article{edstedt2025romav2,
+  title={{RoMa v2: Harder Better Faster Denser Feature Matching}},
+  author={Johan Edstedt, David Nordström, Yushan Zhang, Georg Bökman, Jonathan Astermark, Viktor Larsson, Anders Heyden, Fredrik Kahl, Mårten Wadenbäck, Michael Felsberg},
+  journal={arXiv preprint arXiv:2511.15706},
+  year={2025}
+}
+```

romav2-2.0.0/README.md

@@ -0,0 +1,134 @@
+<p align="center">
+  <h1 align="center"> <ins>RoMa v2</ins> 🤖: Harder Better Faster Denser Feature Matching
+  <h2 align="center">
+    <a href="https://scholar.google.com/citations?user=Ul-vMR0AAAAJ">Johan Edstedt</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=-vJPE04AAAAJ">David Nordström</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=mvY4rdIAAAAJ">Yushan Zhang</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=FUE3Wd0AAAAJ">Georg Bökman</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=dsEPAvUAAAAJ">Jonathan Astermark</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=vHeD0TYAAAAJ">Viktor Larsson</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=9j-6i_oAAAAJ&hl">Anders Heyden</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=P_w6UgMAAAAJ&hl">Fredrik Kahl</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=6WRQpCQAAAAJ">Mårten Wadenbäck</a>
+    ·
+    <a href="https://scholar.google.com/citations?user=lkWfR08AAAAJ">Michael Felsberg</a>
+  </p>
+  <h2 align="center"><p>
+    <a href="https://arxiv.org/abs/2511.15706" align="center">Paper</a> | 
+    <a href="TBD" align="center">Project Page</a>
+  </p></h2>
+  <div align="center"></div>
+</p>
+<br/>
+<p align="center">
+    <img src="assets/qualitative.png" alt="example" width=80%>
+</p>
+
+## How to Use
+```python
+from romav2 import RoMaV2
+
+# load pretrained model
+model = RoMaV2()
+# Match densely for any image-like pair of inputs
+preds = model.match(img_A_path, img_B_path)
+
+# you can also run the forward method directly as 
+# preds = model(img_A, img_B)
+
+# Sample 5000 matches for estimation
+matches, overlaps, precision_AB, precision_BA = model.sample(preds, 5000)
+
+# Convert to pixel coordinates (RoMaV2 produces matches in [-1,1]x[-1,1])
+kptsA, kptsB = model.to_pixel_coordinates(matches, H_A, W_A, H_B, W_B)
+
+# Find a fundamental matrix (or anything else of interest)
+F, mask = cv2.findFundamentalMat(
+    kptsA.cpu().numpy(), kptsB.cpu().numpy(), ransacReprojThreshold=0.2, method=cv2.USAC_MAGSAC, confidence=0.999999, maxIters=10000
+)
+```
+We additionally provide two demos in the [demos folder](demo), which might help in understanding.
+
+
+## Setup/Install
+In your python environment (tested on Linux python 3.12), run:
+```bash
+uv pip install -e .
+```
+or 
+```bash
+uv sync
+```
+
+## Benchmarks
+If you do not already have MegaDepth and ScanNet, you can the following to download them:
+```bash
+source scripts/eval_prep.sh
+```
+### Mega-1500
+```bash
+uv run tests/test_mega1500.py
+```
+### ScanNet-1500
+```bash
+uv run tests/test_scannet1500.py
+```
+### Expected Results
+Experiments on ScanNet-1500 and MegaDepth-1500 are provided in the [tests folder](tests).
+Running these gave me `ScanNet-1500: [34.0, 56.5, 73.9]`, and `Mega-1500: [62.8, 76,8, 86.5]`, which are similar to the results of the paper.
+
+
+## Fused local correlation kernel
+Include the `--extra fused-local-corr` flag as:
+```bash
+uv sync --extra fused-local-corr
+```
+or 
+```bash
+uv pip install romav2[fused-local-corr]
+```
+or
+```bash
+uv add romav2[fused-local-corr]
+```
+
+## Settings
+By twiddling with some different settings you may reach better results on your task of interest.
+Some important ones, which we enable setting to some reasonable defaults through `model.apply_setting`, are:
+
+`model.H_lr, model.W_lr`: height and width for the image pair.
+
+`model.H_hr, model.W_hr`: height and width for a high resolution version of the image pair (used for upsampling as in RoMa)
+
+`model.bidirectional`: Useful for getting more diverse matches, and for estimating the covariance matrix in both directions.
+
+`model.threshold`: Value between [0,1]. Used to set overlap prediction above it to 1. Useful for Mega1500.
+
+`model.balanced_sampling`: Diverse sampling, same as RoMa. Typically helps to get better RANSAC estimates.
+
+## License
+All our code except DINOv3 is MIT license.
+DINOv3 has a custom license, see [DINOv3](https://github.com/facebookresearch/dinov3/tree/main?tab=License-1-ov-file#readme).
+
+## Acknowledgement
+Our codebase builds mainly on the code in [RoMa](https://github.com/Parskatt/RoMa).
+We were additionally inspired by [UFM](https://github.com/UniFlowMatch/UFM) and [MapAnything](https://github.com/facebookresearch/map-anything), particularly for the datasets used to train the models.
+
+## BibTeX
+If you find our models useful, please consider citing our paper!
+```
+@article{edstedt2025romav2,
+  title={{RoMa v2: Harder Better Faster Denser Feature Matching}},
+  author={Johan Edstedt, David Nordström, Yushan Zhang, Georg Bökman, Jonathan Astermark, Viktor Larsson, Anders Heyden, Fredrik Kahl, Mårten Wadenbäck, Michael Felsberg},
+  journal={arXiv preprint arXiv:2511.15706},
+  year={2025}
+}
+```

romav2-2.0.0/pyproject.toml

@@ -0,0 +1,38 @@
+[project]
+name = "romav2"
+version = "2.0.0"
+description = "RoMa v2: Harder Better Faster Denser Feature Matching"
+readme = "README.md"
+authors = [
+    { name = "Johan Edstedt", email = "johan.edstedt@liu.se" }
+]
+requires-python = ">=3.10"
+dependencies = [
+    "dataclasses>=0.8",
+    "einops>=0.8.1",
+    "pillow>=12.0.0",
+    "rich>=14.2.0",
+    "torch",
+    "torchvision>=0.23.0",
+    "tqdm>=4.67.1",
+]
+
+[build-system]
+requires = ["uv_build>=0.8.15,<0.9.0"]
+build-backend = "uv_build"
+
+[project.optional-dependencies]
+fused-local-corr = [
+    "fused-local-corr ; sys_platform == 'linux'",
+]
+eval = [
+    "kornia>=0.8.2",
+    "matplotlib>=3.10.7",
+    "opencv-python>=4.12.0.88",
+    "wandb>=0.23.0",
+    "wxbs-benchmark>=0.0.4",
+]
+dev = [
+    "slurm-util>=0.2.7",
+    "ruff>=0.14.5",
+]

romav2-2.0.0/src/romav2/__init__.py

@@ -0,0 +1,8 @@
+import logging as _logging
+from .logging import configure_logger as configure_logger
+from .logging import logger as _logger
+
+if not any(not isinstance(h, _logging.NullHandler) for h in _logger.handlers):
+    configure_logger()
+
+from .romav2 import RoMaV2 as RoMaV2

romav2-2.0.0/src/romav2/benchmarks/__init__.py

@@ -0,0 +1,4 @@
+from .scannet1500 import ScanNet1500 as ScanNet1500
+from .mega1500 import Mega1500 as Mega1500
+from .wxbs import WxBSBenchmark as WxBSBenchmark
+from .satast import SatAst as SatAst

romav2-2.0.0/src/romav2/benchmarks/mega1500.py

@@ -0,0 +1,115 @@
+import numpy as np
+import logging
+from romav2.geometry import (
+    compute_pose_error,
+    pose_auc,
+    estimate_pose_cv2_ransac,
+    compute_relative_pose,
+)
+from PIL import Image
+from tqdm import tqdm
+
+logger = logging.getLogger(__name__)
+
+
+class Mega1500:
+    def __init__(self, data_root="data/megadepth") -> None:
+        self.scene_names = [
+            "0015_0.1_0.3.npz",
+            "0015_0.3_0.5.npz",
+            "0022_0.1_0.3.npz",
+            "0022_0.3_0.5.npz",
+            "0022_0.5_0.7.npz",
+        ]
+        self.scenes = [
+            np.load(f"{data_root}/{scene}", allow_pickle=True)
+            for scene in self.scene_names
+        ]
+        self.data_root = data_root
+
+    def benchmark(self, model, model_name=None):
+        data_root = self.data_root
+        tot_e_t, tot_e_R, tot_e_pose = [], [], []
+        thresholds = [5, 10, 20]
+        for scene_ind in range(len(self.scenes)):
+            scene = self.scenes[scene_ind]
+            pairs = scene["pair_infos"]
+            intrinsics = scene["intrinsics"]
+            poses = scene["poses"]
+            im_paths = scene["image_paths"]
+            pair_inds = range(len(pairs))
+            for pairind in (pbar := tqdm(pair_inds, desc="Mega1500 eval")):
+                idx1, idx2 = pairs[pairind][0]
+                K1 = intrinsics[idx1].copy()
+                T1 = poses[idx1].copy()
+                R1, t1 = T1[:3, :3], T1[:3, 3]
+                K2 = intrinsics[idx2].copy()
+                T2 = poses[idx2].copy()
+                R2, t2 = T2[:3, :3], T2[:3, 3]
+                R, t = compute_relative_pose(R1, t1, R2, t2)
+                im_A_path = f"{data_root}/{im_paths[idx1]}"
+                im_B_path = f"{data_root}/{im_paths[idx2]}"
+                preds = model.match(im_A_path, im_B_path)
+                im_A = Image.open(im_A_path)
+                w1, h1 = im_A.size
+                im_B = Image.open(im_B_path)
+                w2, h2 = im_B.size
+                if True:  # Note: we keep this true as it was used in DKM/RoMa papers. There is very little difference compared to setting to False.
+                    scale1 = 1200 / max(w1, h1)
+                    scale2 = 1200 / max(w2, h2)
+                    w1, h1 = scale1 * w1, scale1 * h1
+                    w2, h2 = scale2 * w2, scale2 * h2
+                    K1, K2 = K1.copy(), K2.copy()
+                    K1[:2] = K1[:2] * scale1
+                    K2[:2] = K2[:2] * scale2
+                for _ in range(5):
+                    matches, _, _, _ = model.sample(preds, 5_000)
+                    kpts1, kpts2 = model.to_pixel_coordinates(matches, h1, w1, h2, w2)
+                    kpts1, kpts2 = kpts1.cpu().numpy(), kpts2.cpu().numpy()
+                    shuffling = np.random.permutation(np.arange(len(kpts1)))
+                    kpts1 = kpts1[shuffling]
+                    kpts2 = kpts2[shuffling]
+                    try:
+                        threshold = 0.5
+                        norm_threshold = threshold / (
+                            np.mean(np.abs(K1[:2, :2])) + np.mean(np.abs(K2[:2, :2]))
+                        )
+                        R_est, t_est, _ = estimate_pose_cv2_ransac(
+                            kpts1,
+                            kpts2,
+                            K1,
+                            K2,
+                            norm_threshold,
+                            conf=0.99999,
+                        )
+                        e_t, e_R = compute_pose_error(R_est, t_est[:, 0], R, t)
+                        e_pose = max(e_t, e_R)
+                    except Exception as e:
+                        logger.debug(f"Pose estimation error: {e}")
+                        e_t, e_R = 90, 90
+                        e_pose = max(e_t, e_R)
+                    tot_e_t.append(e_t)
+                    tot_e_R.append(e_R)
+                    tot_e_pose.append(e_pose)
+                pbar.set_postfix(
+                    auc=f"{[f'{a.item():.3f}' for a in pose_auc(tot_e_pose, thresholds)]}"
+                )
+
+        tot_e_pose = np.array(tot_e_pose)
+        auc = pose_auc(tot_e_pose, thresholds)
+        acc_5 = (tot_e_pose < 5).mean()
+        acc_10 = (tot_e_pose < 10).mean()
+        acc_15 = (tot_e_pose < 15).mean()
+        acc_20 = (tot_e_pose < 20).mean()
+        map_5 = acc_5
+        map_10 = np.mean([acc_5, acc_10])
+        map_20 = np.mean([acc_5, acc_10, acc_15, acc_20])
+        logger.info("%s auc: %s", model_name, auc)
+        return {
+            "auc_5": auc[0],
+            "auc_10": auc[1],
+            "auc_20": auc[2],
+            "map_5": map_5,
+            "map_10": map_10,
+            "map_20": map_20,
+        }