vision3d 0.1.0__tar.gz

vision3d-0.1.0/LICENSE

@@ -0,0 +1,29 @@
+BSD 3-Clause License
+
+Copyright (c) 2026, Peter Siegel.
+All rights reserved.
+
+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.

vision3d-0.1.0/MANIFEST.in

@@ -0,0 +1,2 @@
+include version.txt
+recursive-include src/vision3d/ops/csrc *.h *.cuh *.cpp *.cu

vision3d-0.1.0/PKG-INFO

@@ -0,0 +1,20 @@
+Metadata-Version: 2.4
+Name: vision3d
+Version: 0.1.0
+Summary: A 3D extension of torchvision.
+Author: Peter Siegel
+License-Expression: BSD-3-Clause
+Project-URL: Homepage, https://github.com/yeetypete/vision3d
+Project-URL: Issues, https://github.com/yeetypete/vision3d/issues
+Project-URL: Repository, https://github.com/yeetypete/vision3d
+Requires-Python: >=3.12
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: torch>=2.10
+Requires-Dist: torchvision>=0.25
+Requires-Dist: typing-extensions>=4.13; python_version < "3.13"
+Provides-Extra: nuscenes
+Requires-Dist: nuscenes-devkit>=1.1.9; extra == "nuscenes"
+Provides-Extra: viz
+Requires-Dist: rerun-sdk>=0.31.1; extra == "viz"
+Dynamic: license-file

vision3d-0.1.0/pyproject.toml

@@ -0,0 +1,82 @@
+[project]
+name = "vision3d"
+description = "A 3D extension of torchvision."
+readme = "README.md"
+requires-python = ">=3.12"
+license = "BSD-3-Clause"
+license-files = ["LICENSE"]
+authors = [{ name = "Peter Siegel" }]
+dependencies = [
+  "torch>=2.10",
+  "torchvision>=0.25",
+  "typing-extensions>=4.13; python_version < '3.13'",
+]
+dynamic = ["version"]
+
+[project.urls]
+Homepage = "https://github.com/yeetypete/vision3d"
+Issues = "https://github.com/yeetypete/vision3d/issues"
+Repository = "https://github.com/yeetypete/vision3d"
+
+[project.optional-dependencies]
+nuscenes = [
+  "nuscenes-devkit>=1.1.9",
+]
+viz = [
+  "rerun-sdk>=0.31.1",
+]
+
+[dependency-groups]
+dev = [
+  "pyrefly>=0.59.1",
+  "pytest>=9.0.2",
+  "ruff>=0.15.9",
+]
+
+[build-system]
+requires = ["numpy>=2", "setuptools<82", "torch>=2.10"]
+build-backend = "setuptools.build_meta"
+
+[tool.pyrefly]
+min-severity = "ignore"
+search-path = ["examples", "test"]
+
+[tool.pyrefly.errors]
+# Pyrefly does not yet have a strict mode, so we manually set all diagnostics
+# with "ignore" severity levels to "warn".
+implicit-abstract-class = "warn"
+implicit-any = "warn"
+implicitly-defined-attribute = "warn"
+missing-override-decorator = "warn"
+missing-source = "warn"
+not-required-key-access = "warn"
+open-unpacking = "warn"
+unannotated-attribute = "warn"
+unannotated-parameter = "warn"
+unannotated-return = "warn"
+untyped-import = "warn"
+unused-ignore = "warn"
+
+[tool.ruff.lint]
+preview = true
+extend-select = ["D", "DOC", "PT", "TC"]
+ignore = [
+  "D104",  # TODO: remove, missing docstring in placeholder __init__.py packages
+  "D107",  # __init__ args documented in class docstring (Google style)
+]
+
+[tool.ruff.lint.per-file-ignores]
+"test/**" = ["D"]
+"examples/**" = ["D"]
+
+[tool.ruff.lint.pydocstyle]
+convention = "google"
+
+[tool.ruff.format]
+docstring-code-format = true
+
+[tool.pytest.ini_options]
+filterwarnings = [
+  # Unrelated warning from ``@torch.jit.script_method`` in torch/utils/mkldnn.py.
+  "ignore:.*torch\\.jit\\.script_method.*:DeprecationWarning",
+]

vision3d-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

vision3d-0.1.0/setup.py

@@ -0,0 +1,83 @@
+"""setuptools entry point for vision3d."""
+
+import os
+import subprocess
+from pathlib import Path
+from typing import override
+
+import torch
+from setuptools import setup
+from setuptools.command.sdist import sdist
+from torch.utils.cpp_extension import (
+    CUDA_HOME,
+    BuildExtension,
+    CppExtension,
+    CUDAExtension,
+)
+
+_ROOT = Path(__file__).resolve().parent
+
+
+def get_version() -> str:
+    """Return the project version.
+
+    If the ``BUILD_VERSION`` environment variable is set, it fully overrides the
+    base version read from ``version.txt``. Otherwise, for local builds the
+    the short git commit SHA is appened as a PEP 440 local version identifier.
+    """
+    if build_version := os.getenv("BUILD_VERSION"):
+        return build_version
+
+    with open(_ROOT / "version.txt") as f:
+        version = f.readline().strip()
+
+    if "+" in version:
+        return version
+
+    try:
+        sha = (
+            subprocess.check_output(["git", "rev-parse", "HEAD"], cwd=str(_ROOT))
+            .decode("ascii")
+            .strip()
+        )
+    except (subprocess.CalledProcessError, FileNotFoundError):
+        return version
+
+    return f"{version}+{sha[:7]}"
+
+
+class VersionedSdist(sdist):
+    """Bake the fully-resolved version into the sdist's ``version.txt``."""
+
+    @override
+    def make_release_tree(self, base_dir: str, files: list[str]) -> None:
+        """Write the resolved version into ``version.txt`` in the release tree."""
+        super().make_release_tree(base_dir, files)
+        (Path(base_dir) / "version.txt").write_text(f"{get_version()}\n")
+
+
+FORCE_CUDA = os.getenv("FORCE_CUDA", "0") == "1"
+_HAS_CUDA = (torch.cuda.is_available() and CUDA_HOME is not None) or FORCE_CUDA
+
+_CSRC = _ROOT / "src/vision3d/ops/csrc"
+_SOURCES = [
+    "src/vision3d/ops/csrc/iou_box3d.cpp",
+    "src/vision3d/ops/csrc/iou_box3d/iou_box3d_cpu.cpp",
+]
+if _HAS_CUDA:
+    _SOURCES.append("src/vision3d/ops/csrc/iou_box3d/iou_box3d.cu")
+
+Extension = CUDAExtension if _HAS_CUDA else CppExtension
+
+setup(
+    version=get_version(),
+    ext_modules=[
+        Extension(
+            name="vision3d._C",
+            sources=_SOURCES,
+            include_dirs=[str(_CSRC)],
+            define_macros=[("WITH_CUDA", None)] if _HAS_CUDA else [],
+        ),
+    ],
+    cmdclass={"build_ext": BuildExtension, "sdist": VersionedSdist},
+)

vision3d-0.1.0/src/vision3d/_extension.py

@@ -0,0 +1,30 @@
+"""Load vision3d's compiled C++/CUDA extension.
+
+Importing this module has the side effect of loading the ops into the dispatcher.
+"""
+
+import importlib.machinery
+import os
+
+import torch
+
+
+def _get_extension_path(lib_name: str) -> str:
+    lib_dir = os.path.dirname(__file__)
+    loader_details = (
+        importlib.machinery.ExtensionFileLoader,
+        importlib.machinery.EXTENSION_SUFFIXES,
+    )
+    extfinder = importlib.machinery.FileFinder(lib_dir, loader_details)
+    ext_specs = extfinder.find_spec(lib_name)
+    if ext_specs is None or ext_specs.origin is None:
+        msg = f"Could not find module '{lib_name}' in {lib_dir}."
+        raise ImportError(msg)
+    return ext_specs.origin
+
+
+def _load_library(lib_name: str) -> None:
+    torch.ops.load_library(_get_extension_path(lib_name))
+
+
+_load_library("_C")

vision3d-0.1.0/src/vision3d/datasets/__init__.py

@@ -0,0 +1,21 @@
+from ._types import (
+    CameraInputs,
+    FusionInputs,
+    LidarInputs,
+    SampleInputs,
+    SampleTargets,
+)
+from .collate import collate_fn
+from .kitti import Kitti3D
+from .nuscenes import NuScenes3D
+
+__all__ = [
+    "CameraInputs",
+    "FusionInputs",
+    "Kitti3D",
+    "LidarInputs",
+    "NuScenes3D",
+    "SampleInputs",
+    "SampleTargets",
+    "collate_fn",
+]

vision3d-0.1.0/src/vision3d/datasets/_types.py

@@ -0,0 +1,73 @@
+"""Types for :mod:`vision3d.datasets` samples."""
+
+import sys
+from typing import NotRequired, Required, TypedDict
+
+from torch import Tensor
+
+if sys.version_info >= (3, 13):
+    from typing import ReadOnly
+else:
+    from typing_extensions import ReadOnly
+
+from vision3d.tensors import (
+    BoundingBoxes3D,
+    CameraExtrinsics,
+    CameraImages,
+    CameraIntrinsics,
+    PointCloud3D,
+)
+
+
+class SampleInputs(TypedDict):
+    """Per-frame model inputs; base contract with all fields optional.
+
+    Fields are :data:`ReadOnly` so dataset-specific subclasses can
+    tighten them from :data:`NotRequired` to :data:`Required`.
+
+    Attributes:
+        points: Lidar point cloud for the frame.
+        images: Multi-camera image tensor, one row per camera.
+        extrinsics: Lidar-to-camera transforms, one row per camera.
+        intrinsics: Per-camera pinhole intrinsic matrices.
+    """
+
+    points: NotRequired[ReadOnly[PointCloud3D]]
+    images: NotRequired[ReadOnly[CameraImages]]
+    extrinsics: NotRequired[ReadOnly[CameraExtrinsics]]
+    intrinsics: NotRequired[ReadOnly[CameraIntrinsics]]
+
+
+class LidarInputs(SampleInputs):
+    """Lidar-only sample: points always present."""
+
+    points: Required[PointCloud3D]
+
+
+class CameraInputs(SampleInputs):
+    """Camera-only sample: images, intrinsics, and extrinsics always present."""
+
+    images: Required[CameraImages]
+    intrinsics: Required[CameraIntrinsics]
+    extrinsics: Required[CameraExtrinsics]
+
+
+class FusionInputs(SampleInputs):
+    """Fusion sample: lidar plus multi-camera, every field present."""
+
+    points: Required[PointCloud3D]
+    images: Required[CameraImages]
+    extrinsics: Required[CameraExtrinsics]
+    intrinsics: Required[CameraIntrinsics]
+
+
+class SampleTargets(TypedDict):
+    """Per-frame ground-truth annotations.
+
+    Attributes:
+        boxes: 3D bounding boxes in the lidar frame.
+        labels: Integer class labels, one per box.
+    """
+
+    boxes: BoundingBoxes3D
+    labels: Tensor

vision3d-0.1.0/src/vision3d/datasets/collate.py

@@ -0,0 +1,22 @@
+"""Collation utilities for DataLoader."""
+
+from typing import Any
+
+
+def collate_fn(
+    batch: list[tuple[Any, Any]],
+) -> tuple[tuple[Any, ...], tuple[Any, ...]]:
+    """Collate a batch of ``(inputs, targets)`` without stacking.
+
+    Variable-size tensors (point clouds, bounding boxes) cannot be stacked
+    into a single tensor. This collate function groups them as tuples,
+    matching torchvision's detection collate pattern.
+
+    Args:
+        batch: List of ``(inputs, targets)`` from the dataset.
+
+    Returns:
+        Tuple of ``(inputs_tuple, targets_tuple)``.
+    """
+    inputs, targets = zip(*batch)
+    return tuple(inputs), tuple(targets)