libviper 0.1.0__tar.gz

libviper-0.1.0/.gitignore

@@ -0,0 +1,165 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# env file
+.env

libviper-0.1.0/LICENSE

@@ -0,0 +1,28 @@
+BSD 3-Clause License
+
+Copyright (c) 2025, Martin Kvisvik Larsen
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+1. Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+2. Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+3. Neither the name of the copyright holder nor the names of its
+   contributors may be used to endorse or promote products derived from
+   this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

libviper-0.1.0/PKG-INFO

@@ -0,0 +1,165 @@
+Metadata-Version: 2.4
+Name: libviper
+Version: 0.1.0
+Summary: A package for visual place recognition (VPR) models.
+Author-email: Martin Kvisvik Larsen <martin.kvisvik.larsen@hotmail.com>
+License-File: LICENSE
+Requires-Python: >=3.12
+Requires-Dist: einops>=0.8.1
+Requires-Dist: fast-pytorch-kmeans>=0.2.2
+Requires-Dist: imageio>=2.37.0
+Requires-Dist: loguru>=0.7.3
+Requires-Dist: msgspec>=0.19.0
+Requires-Dist: pytest>=8.3.4
+Requires-Dist: python-dotenv>=1.0.1
+Requires-Dist: pytorch-lightning>=2.6.0
+Requires-Dist: ruff>=0.8.3
+Requires-Dist: scipy>=1.16.0
+Requires-Dist: torch>=2.7.0
+Requires-Dist: torchvision>=0.24.0
+Requires-Dist: tqdm>=4.67.1
+Requires-Dist: xformers>=0.0.30
+Description-Content-Type: text/markdown
+
+# Viper: A Common Image Embedder Interface for Visual Place Recognition
+
+![ci](https://github.com/markvilar/viper/actions/workflows/ubuntu.yml/badge.svg)
+
+This Python package provides a **unified** image embedder interface for visual place recognition (VPR), along with wrapper implementations of several state-of-the-art VPR models so they all expose the same API.
+It also includes a lightweight registry mechanism that lets you register custom embedders and retrieve them by string key.
+
+## Features
+
+- Common `ImageEmbedder` protocol for VPR models (name, vector size, device, call semantics).
+- Eight wrapper models adapting popular VPR architectures to this interface:
+    - AnyLoc
+    - CliqueMining
+    - CosPlace
+    - EigenPlaces
+    - MegaLoc
+    - MixVPR
+    - NetVLAD
+    - SALAD
+- Simple registry (`register_embedder_factory` / `get_embedder_factory`) for instantiating embedders by key.
+
+## Installation
+
+The package is configured as a standard Python project via `pyproject.toml` and adds support for the `uv` package manager.
+
+You can install it in editable mode for development:
+
+```bash
+uv sync
+```
+
+**Disclaimer:** AnyLoc, CliqueMining, MegaLoc, and SALAD require CUDA to run.
+
+## Usage
+
+### Loading a built-in embedder
+
+The recommended way to construct models is through the embedder registry.
+
+```python
+import viper
+
+factory: viper.ImageEmbedderFactory = viper.get_embedder_factory("salad")    # or "mixvpr", "netvlad", "eigenplaces", ...
+embedder: viper.ImageEmbedder = factory()
+
+print(embedder.name)                   # "salad"
+print(embedder.vector_size)            # 8448
+print(embedder.embedder_parameters)    # dict of parameters
+print(embedder.device)                 # "cpu" or "cuda"
+```
+
+All embedders implement the `ImageEmbedder` protocol, which exposes:
+
+- `name: str`
+- `vector_size: int` (embedding dimension)
+- `embedder_parameters: dict[str, Any]` (model-specific metadata such as backbone, descriptor size, etc.)
+- `device: str` (e.g. `"cpu"` or `"cuda:0"`)
+- `__call__(images: torch.Tensor) -> torch.Tensor` (batched embedding, `B x C x H x W -> B x E`)
+
+
+### Embedding a batch of images
+
+All wrappers expect a batch of images as a float tensor of shape `B x C x H x W` with values in `[0.0, 1.0]` (for NetVLAD this is explicitly asserted).
+
+```python
+import torch
+import viper
+
+factory: viper.ImageEmbedderFactory = viper.get_embedder_factory("netvlad")
+embedder: viper.ImageEmbedder = factory()
+images: torch.Tensor = torch.rand(8, 3, 480, 640)  # example batch, normalized to [0, 1]
+embeddings: torch.Tensor = embedder(images)        # shape: (8, embedder.vectorsize)
+```
+
+Wrappers handle grayscale input by converting 1-channel batches to 3-channel RGB internally.
+Some models also resize images so that height/width are multiples of 14, matching DINOv2 backbone constraints.
+
+### Registering a custom model
+
+You can register your own model as long as it fulfills the `ImageEmbedder` interface.
+
+```python
+import torch
+from viper.registry import register_embedder_factory
+from viper.types import ImageEmbedder
+
+class MyEmbedder(torch.nn.Module):
+    @property
+    def name(self) -> str:
+        return "myembedder"
+
+    @property
+    def vector_size(self) -> int:
+        return 1024
+
+    @property
+    def embedder_parameters(self) -> dict:
+        return {"backbone": "resnet50"}
+
+    @property
+    def device(self) -> str:
+        return next(self.parameters()).device.type
+
+    def forward(self, images: torch.Tensor) -> torch.Tensor:
+        # return embeddings of shape B x 1024
+        ...
+
+# Factory that returns an ImageEmbedder instance
+@register_embedder_factory(key="myembedder")
+def load_myembedder() -> ImageEmbedder:
+    model = MyEmbedder()
+    return model
+
+```
+
+You can then retrieve a factory via `get_embedder_factory("myembedder")` like the built-in models.
+
+
+## Testing
+
+The repository includes tests for the registry and embedder factories.[^10]
+
+To run them:
+
+```bash
+uv run pytest
+```
+
+
+## Acknowledgements
+
+This package reuses ideas, code, and checkpoints from several excellent VPR projects.
+Please cite and credit the original works when using the corresponding models.
+
+- [AnyLoc](https://github.com/AnyLoc/DINO)
+- [CliqueMining](https://github.com/serizba/clique-mining)
+- [CosPlace](https://github.com/gmberton/CosPlace)
+- [EigenPlaces](https://github.com/gmberton/EigenPlaces)
+- [MegaLoc](https://github.com/gmberton/MegaLoc)
+- [MixVPR](https://github.com/amaralibey/MixVPR)
+- [NetVLAD](https://github.com/cvg/Hierarchical-Localization)
+- [SALAD](https://github.com/serizba/salad)

libviper-0.1.0/README.md

@@ -0,0 +1,142 @@
+# Viper: A Common Image Embedder Interface for Visual Place Recognition
+
+![ci](https://github.com/markvilar/viper/actions/workflows/ubuntu.yml/badge.svg)
+
+This Python package provides a **unified** image embedder interface for visual place recognition (VPR), along with wrapper implementations of several state-of-the-art VPR models so they all expose the same API.
+It also includes a lightweight registry mechanism that lets you register custom embedders and retrieve them by string key.
+
+## Features
+
+- Common `ImageEmbedder` protocol for VPR models (name, vector size, device, call semantics).
+- Eight wrapper models adapting popular VPR architectures to this interface:
+    - AnyLoc
+    - CliqueMining
+    - CosPlace
+    - EigenPlaces
+    - MegaLoc
+    - MixVPR
+    - NetVLAD
+    - SALAD
+- Simple registry (`register_embedder_factory` / `get_embedder_factory`) for instantiating embedders by key.
+
+## Installation
+
+The package is configured as a standard Python project via `pyproject.toml` and adds support for the `uv` package manager.
+
+You can install it in editable mode for development:
+
+```bash
+uv sync
+```
+
+**Disclaimer:** AnyLoc, CliqueMining, MegaLoc, and SALAD require CUDA to run.
+
+## Usage
+
+### Loading a built-in embedder
+
+The recommended way to construct models is through the embedder registry.
+
+```python
+import viper
+
+factory: viper.ImageEmbedderFactory = viper.get_embedder_factory("salad")    # or "mixvpr", "netvlad", "eigenplaces", ...
+embedder: viper.ImageEmbedder = factory()
+
+print(embedder.name)                   # "salad"
+print(embedder.vector_size)            # 8448
+print(embedder.embedder_parameters)    # dict of parameters
+print(embedder.device)                 # "cpu" or "cuda"
+```
+
+All embedders implement the `ImageEmbedder` protocol, which exposes:
+
+- `name: str`
+- `vector_size: int` (embedding dimension)
+- `embedder_parameters: dict[str, Any]` (model-specific metadata such as backbone, descriptor size, etc.)
+- `device: str` (e.g. `"cpu"` or `"cuda:0"`)
+- `__call__(images: torch.Tensor) -> torch.Tensor` (batched embedding, `B x C x H x W -> B x E`)
+
+
+### Embedding a batch of images
+
+All wrappers expect a batch of images as a float tensor of shape `B x C x H x W` with values in `[0.0, 1.0]` (for NetVLAD this is explicitly asserted).
+
+```python
+import torch
+import viper
+
+factory: viper.ImageEmbedderFactory = viper.get_embedder_factory("netvlad")
+embedder: viper.ImageEmbedder = factory()
+images: torch.Tensor = torch.rand(8, 3, 480, 640)  # example batch, normalized to [0, 1]
+embeddings: torch.Tensor = embedder(images)        # shape: (8, embedder.vectorsize)
+```
+
+Wrappers handle grayscale input by converting 1-channel batches to 3-channel RGB internally.
+Some models also resize images so that height/width are multiples of 14, matching DINOv2 backbone constraints.
+
+### Registering a custom model
+
+You can register your own model as long as it fulfills the `ImageEmbedder` interface.
+
+```python
+import torch
+from viper.registry import register_embedder_factory
+from viper.types import ImageEmbedder
+
+class MyEmbedder(torch.nn.Module):
+    @property
+    def name(self) -> str:
+        return "myembedder"
+
+    @property
+    def vector_size(self) -> int:
+        return 1024
+
+    @property
+    def embedder_parameters(self) -> dict:
+        return {"backbone": "resnet50"}
+
+    @property
+    def device(self) -> str:
+        return next(self.parameters()).device.type
+
+    def forward(self, images: torch.Tensor) -> torch.Tensor:
+        # return embeddings of shape B x 1024
+        ...
+
+# Factory that returns an ImageEmbedder instance
+@register_embedder_factory(key="myembedder")
+def load_myembedder() -> ImageEmbedder:
+    model = MyEmbedder()
+    return model
+
+```
+
+You can then retrieve a factory via `get_embedder_factory("myembedder")` like the built-in models.
+
+
+## Testing
+
+The repository includes tests for the registry and embedder factories.[^10]
+
+To run them:
+
+```bash
+uv run pytest
+```
+
+
+## Acknowledgements
+
+This package reuses ideas, code, and checkpoints from several excellent VPR projects.
+Please cite and credit the original works when using the corresponding models.
+
+- [AnyLoc](https://github.com/AnyLoc/DINO)
+- [CliqueMining](https://github.com/serizba/clique-mining)
+- [CosPlace](https://github.com/gmberton/CosPlace)
+- [EigenPlaces](https://github.com/gmberton/EigenPlaces)
+- [MegaLoc](https://github.com/gmberton/MegaLoc)
+- [MixVPR](https://github.com/amaralibey/MixVPR)
+- [NetVLAD](https://github.com/cvg/Hierarchical-Localization)
+- [SALAD](https://github.com/serizba/salad)

libviper-0.1.0/pyproject.toml

@@ -0,0 +1,50 @@
+[project]
+name = "libviper"
+version = "0.1.0"
+description = "A package for visual place recognition (VPR) models."
+readme = "README.md"
+authors = [
+  { name = "Martin Kvisvik Larsen", email = "martin.kvisvik.larsen@hotmail.com" }
+]
+requires-python = ">=3.12"
+dependencies = [
+  "imageio>=2.37.0",
+  "loguru>=0.7.3",
+  "msgspec>=0.19.0",
+  "pytest>=8.3.4",
+  "python-dotenv>=1.0.1",
+  "ruff>=0.8.3",
+  "torch>=2.7.0",
+  "torchvision>=0.24.0",
+  "tqdm>=4.67.1",
+  "scipy>=1.16.0",
+  "einops>=0.8.1", # NOTE: required by anyloc
+  "xformers>=0.0.30", # NOTE: required by megaloc
+  "pytorch-lightning>=2.6.0", # NOTE: required by cliquemining
+  "fast-pytorch-kmeans>=0.2.2", # NOTE: required by anyloc
+]
+
+[project.scripts]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build]
+include = ["src/viper"]
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/viper"]
+
+[tool.uv]
+package = true
+
+[tool.uv.sources]
+
+[tool.ruff]
+line-length = 88 # default line width of ruff format
+
+[dependency-groups]
+dev = [
+  "ruff>=0.8.3",
+]

libviper-0.1.0/src/viper/__init__.py

@@ -0,0 +1,15 @@
+"""
+Package for visual place recognition (VPR) models.
+"""
+
+import viper.models as models  # noqa: F401
+
+from .registry import FactoryRegistry as FactoryRegistry
+from .registry import register_embedder_factory as register_embedder_factory
+from .registry import get_embedder_factory_registry as get_embedder_factory_registry
+from .registry import get_embedder_factory as get_embedder_factory
+
+from .types import ImageEmbedder as ImageEmbedder
+from .types import ImageEmbedderFactory as ImageEmbedderFactory
+
+__all__ = []

libviper-0.1.0/src/viper/models/__init__.py

@@ -0,0 +1,15 @@
+"""
+Package with vipers builtin VPR models.
+"""
+
+import viper.models.anyloc as anyloc  # noqa: F401
+import viper.models.cliquemining as cliquemining  # noqa: F401
+import viper.models.cosplace as cosplace  # noqa: F401
+import viper.models.eigenplaces as eigenplaces  # noqa: F401
+import viper.models.megaloc as megaloc  # noqa: F401
+import viper.models.mixvpr as mixvpr  # noqa: F401
+import viper.models.netvlad as netvlad  # noqa: F401
+import viper.models.salad as salad  # noqa: F401
+
+
+__all__ = []

libviper-0.1.0/src/viper/models/anyloc.py

@@ -0,0 +1,92 @@
+"""Module for the Anyloc VPR model."""
+
+import typing
+import torch
+
+from viper.registry import register_embedder_factory
+from viper.types import ImageEmbedder
+from .helpers import convert_grayscale_batch_to_rgb
+from .helpers import calculate_image_size_dinov2
+from .helpers import resize_image_batch
+
+
+class AnyLocWrapper(torch.nn.Module):
+    """
+    Class representing a wrapper for the AnyLoc model.
+    """
+
+    def __init__(self, impl: torch.nn.Module) -> None:
+        """Initializer method."""
+        super().__init__()
+        self._impl = impl
+        # NOTE: Add dummy parameter to infer device
+        self._param = torch.nn.Parameter(torch.tensor(1.0))
+
+    @property
+    def name(self) -> str:
+        """Returns the name of the embedder."""
+        return "anyloc"
+
+    @property
+    def vector_size(self) -> int:
+        """Returns the size, i.e. dimensions, of the image embeddings."""
+        return self._impl.vlad.num_clusters * self._impl.vlad.desc_dim
+
+    @property
+    def embedder_parameters(self) -> dict[str, typing.Any]:
+        """Returns the parameters of the embedder."""
+        return {
+            "num_clusters": self._impl.vlad.num_clusters,
+            "descriptor_dimensions": self._impl.vlad.desc_dim,
+        }
+
+    @property
+    def device(self) -> str:
+        """Returns the device of the embedder."""
+        return next(self.parameters()).device
+
+    def __call__(self, images: torch.Tensor) -> torch.Tensor:
+        """
+        Embeddes a batch of images.
+        :arg images: tensor of shape BxCxHxW
+        """
+        return self.forward(images)
+
+    def forward(self, images: torch.Tensor) -> torch.Tensor:
+        """
+        Forwards a batch of images through the module.
+
+        Arguments:
+            images: batch of images - shape BxCxHxW
+        Returns:
+            batch of image embeddings - shape BxE
+        """
+        # If the image batch is grayscale, convert to 3 channels
+        if images.shape[1] == 1:
+            images: torch.Tensor = convert_grayscale_batch_to_rgb(images)
+
+        assert images.dim() == 4, f"invalid batch dimensions: {images.dim()}"
+        assert images.shape[1] == 3, f"invalid image batch channels: {images.shape[1]}"
+
+        desired_image_size: tuple[int, int] = calculate_image_size_dinov2(images)
+        images_resized: torch.Tensor = resize_image_batch(images, desired_image_size)
+
+        return self._impl(images_resized)
+
+
+@register_embedder_factory(key="anyloc")
+def load_anyloc() -> ImageEmbedder:
+    """Loads an AnyLoc model."""
+    # NOTE: AnyLoc requires CUDA to run, hence we assert
+    if not torch.cuda.is_available():
+        raise RuntimeError("CUDA is required for AnyLoc but is not available.")
+
+    impl: torch.nn.Module = torch.hub.load(
+        "AnyLoc/DINO",
+        "get_vlad_model",
+        backbone="DINOv2",
+        domain="unstructured",
+        device="cuda",
+    )
+    wrapper: AnyLocWrapper = AnyLocWrapper(impl=impl).eval()
+    return wrapper