tarsius 0.1.0__py3-none-any.whl

tarsius/__init__.py

@@ -0,0 +1,41 @@
+"""Local, training-free state detection for a region of a camera image."""
+
+from importlib.metadata import version
+
+from .backends.edges import EdgeBackend
+from .backends.ncnn import NcnnBackend
+from .classifier import Classifier, Result, Status
+from .errors import (
+    BackendError,
+    BackendMismatchError,
+    FeaturesError,
+    ImageError,
+    TarsiusError,
+)
+from .features import FeatureBackend, Features, Reference, extract_features
+from .image import load_image
+from .roi import FULL_FRAME, Roi
+from .vote import majority
+
+__version__ = version("tarsius")
+
+__all__ = [
+    "FULL_FRAME",
+    "BackendError",
+    "BackendMismatchError",
+    "Classifier",
+    "EdgeBackend",
+    "FeatureBackend",
+    "Features",
+    "FeaturesError",
+    "ImageError",
+    "NcnnBackend",
+    "Reference",
+    "Result",
+    "Roi",
+    "Status",
+    "TarsiusError",
+    "extract_features",
+    "load_image",
+    "majority",
+]

tarsius/backends/__init__.py

@@ -0,0 +1 @@
+"""Feature backends."""

tarsius/backends/edges.py

@@ -0,0 +1,49 @@
+"""Edge map backend: numpy only, needs a tight region."""
+
+from __future__ import annotations
+
+import numpy as np
+from PIL import Image, ImageFilter
+
+from ..features import Vector, unit
+
+SIZE = (128, 96)
+# Pixels the region may move between frames, e.g. after a camera bump.
+SHIFT = 2
+
+
+class EdgeBackend:
+    """
+    Gradient magnitude compared by normalized cross-correlation.
+
+    Baseline without extra dependencies. Anything moved inside the region
+    changes the edges, so the region must be tight, and it must keep some
+    static structure (door, wall edges) for the empty state: a featureless
+    region has nothing to compare.
+    """
+
+    id = "edges/1"
+    default_margin = 0.02
+
+    def extract(self, image: Image.Image) -> tuple[Vector, Vector]:
+        """Return the centre crop and every shifted window of the edge map."""
+        gray = image.convert("L").resize(SIZE, Image.Resampling.BILINEAR)
+        pixels = np.asarray(gray.filter(ImageFilter.GaussianBlur(1)), np.float32)
+        grad_x = np.zeros_like(pixels)
+        grad_y = np.zeros_like(pixels)
+        grad_x[:, 1:-1] = pixels[:, 2:] - pixels[:, :-2]
+        grad_y[1:-1, :] = pixels[2:, :] - pixels[:-2, :]
+        edges = np.hypot(grad_x, grad_y)
+        height, width = edges.shape
+        inner = (slice(SHIFT, height - SHIFT), slice(SHIFT, width - SHIFT))
+        windows = [
+            unit(
+                edges[
+                    SHIFT + dy : height - SHIFT + dy,
+                    SHIFT + dx : width - SHIFT + dx,
+                ]
+            )
+            for dy in range(-SHIFT, SHIFT + 1)
+            for dx in range(-SHIFT, SHIFT + 1)
+        ]
+        return unit(edges[inner]), np.stack(windows)

tarsius/backends/ncnn.py

@@ -0,0 +1,76 @@
+"""SqueezeNet embedding backend on ncnn."""
+
+from __future__ import annotations
+
+from importlib.resources import as_file, files
+
+import numpy as np
+from PIL import Image
+
+from ..errors import BackendError
+from ..features import Vector
+
+_INPUT = 227
+_MEAN = [104.0, 117.0, 123.0]
+# Last feature map before the ImageNet classifier, 512 channels.
+_BLOB = "fire9/concat_drop9"
+
+
+class NcnnBackend:
+    """
+    ImageNet SqueezeNet 1.1 features, averaged over the feature map.
+
+    No training: the network only describes the region, the classifier
+    compares those descriptions. Robust to clutter and lighting, so a rough
+    region is enough.
+    """
+
+    # Bump with any change to the model, ncnn options or preprocessing.
+    id = "squeezenet11/1"
+    default_margin = 0.02
+
+    def __init__(self, *, threads: int = 1) -> None:
+        """Load the bundled model; raises ImportError without the ``ncnn`` extra."""
+        try:
+            import ncnn  # noqa: PLC0415
+        except ImportError as err:  # pragma: no cover - depends on the environment
+            raise ImportError("NcnnBackend needs `pip install tarsius[ncnn]`") from err
+        self._ncnn = ncnn
+        self._net = ncnn.Net()
+        opt = self._net.opt
+        opt.num_threads = threads
+        # fp16 arithmetic differs between CPUs; keep features comparable when
+        # references move to other hardware, e.g. restored from a backup.
+        opt.use_fp16_arithmetic = False
+        opt.use_vulkan_compute = False
+        models = files("tarsius") / "models"
+        param = models / "squeezenet_v1.1.param"
+        model = models / "squeezenet_v1.1.bin"
+        if not (param.is_file() and model.is_file()):
+            raise BackendError("bundled SqueezeNet model is missing")
+        with as_file(param) as param_path, as_file(model) as model_path:
+            if self._net.load_param(str(param_path)) or self._net.load_model(
+                str(model_path)
+            ):
+                raise BackendError("ncnn could not load the SqueezeNet model")
+
+    def extract(self, image: Image.Image) -> tuple[Vector, Vector]:
+        """Return the unit-norm embedding as both reference and query."""
+        ncnn = self._ncnn
+        # PIL antialiases when shrinking, ncnn's own resize does not.
+        resized = image.convert("RGB").resize(
+            (_INPUT, _INPUT), Image.Resampling.BILINEAR
+        )
+        bgr = np.ascontiguousarray(np.asarray(resized)[..., ::-1])
+        mat = ncnn.Mat.from_pixels(bgr, ncnn.Mat.PixelType.PIXEL_BGR, _INPUT, _INPUT)
+        mat.substract_mean_normalize(_MEAN, [])
+        extractor = self._net.create_extractor()
+        extractor.input("data", mat)
+        ret, out = extractor.extract(_BLOB)
+        if ret:  # pragma: no cover - ncnn reports no recoverable cause
+            raise BackendError(f"ncnn extraction failed with {ret}")
+        maps = np.array(out, dtype=np.float32)
+        embedding = maps.reshape(maps.shape[0], -1).mean(axis=1)
+        norm = float(np.linalg.norm(embedding))
+        vector = embedding / norm if norm > 0 else embedding
+        return vector, vector[None]

tarsius/classifier.py

@@ -0,0 +1,179 @@
+"""Nearest-reference classifier with explicit "unsure" and "not ready" results."""
+
+from __future__ import annotations
+
+import threading
+from collections.abc import Mapping
+from dataclasses import dataclass, field
+from enum import StrEnum
+from typing import NamedTuple
+
+import numpy as np
+import numpy.typing as npt
+
+from .errors import BackendMismatchError
+from .features import FeatureBackend, Features, Reference, Vector
+
+
+class Status(StrEnum):
+    """Why a result has or lacks a state."""
+
+    CONFIDENT = "confident"
+    # The best state does not beat the runner-up by the margin: ask the user.
+    UNSURE = "unsure"
+    # Fewer than two states have references: nothing to compare yet.
+    NOT_READY = "not_ready"
+
+
+@dataclass(frozen=True, slots=True)
+class Result:
+    """Outcome of one classification."""
+
+    status: Status
+    # Set only when confident.
+    state: str | None = None
+    # Best state even when unsure, e.g. to ask "is this <candidate>?".
+    candidate: str | None = None
+    margin: float = 0.0
+    # Diagnostics only; the scale depends on the backend.
+    scores: Mapping[str, float] = field(default_factory=dict)
+    # Most similar reference key per state.
+    nearest: Mapping[str, str] = field(default_factory=dict)
+
+
+class _Index(NamedTuple):
+    keys: tuple[str, ...]
+    labels: npt.NDArray[np.str_]
+    matrix: Vector
+
+
+class Classifier:
+    """
+    k-nearest references per state, compared by similarity.
+
+    A state wins only when its mean top-k similarity beats every other state
+    by ``margin``. ``k`` shrinks to the smallest state's reference count, so
+    adding references to one state never makes it lose. Safe to share between
+    threads: changes swap in a new snapshot that lookups read once.
+    """
+
+    def __init__(self, backend_id: str, *, margin: float, k: int = 3) -> None:
+        """Create an empty classifier for features of ``backend_id``."""
+        if k < 1:
+            raise ValueError("k must be at least 1")
+        if margin <= 0:
+            raise ValueError("margin must be positive, a tie is never confident")
+        self.backend_id = backend_id
+        self.margin = margin
+        self.k = k
+        self._lock = threading.Lock()
+        self._refs: dict[str, tuple[str, Vector]] = {}
+        self._index: _Index | None = None
+
+    @classmethod
+    def for_backend(cls, backend: FeatureBackend, *, k: int = 3) -> Classifier:
+        """Create a classifier with the backend's id and default margin."""
+        return cls(backend.id, margin=backend.default_margin, k=k)
+
+    def add(self, key: str, features: Features | Reference, *, state: str) -> None:
+        """
+        Add or replace the reference ``key``.
+
+        Raises ValueError for a region without any structure, e.g. a blank
+        frame from an offline camera, since it would match nothing.
+        """
+        self._check(features.backend_id)
+        vector = features.vector
+        if not np.linalg.norm(vector) > 0:
+            raise ValueError("reference has no structure to compare")
+        with self._lock:
+            index = self._index
+            if index is not None and index.matrix.shape[1] != vector.shape[0]:
+                raise BackendMismatchError(
+                    f"vector length {vector.shape[0]}, expected {index.matrix.shape[1]}"
+                )
+            self._refs[key] = (state, vector)
+            self._index = self._build()
+
+    def remove(self, key: str) -> None:
+        """Drop the reference ``key`` if present."""
+        with self._lock:
+            if self._refs.pop(key, None) is not None:
+                self._index = self._build()
+
+    def clear(self) -> None:
+        """Drop all references."""
+        with self._lock:
+            self._refs.clear()
+            self._index = None
+
+    def references(self) -> dict[str, str]:
+        """Return the state of every reference key."""
+        index = self._index
+        if index is None:
+            return {}
+        return dict(zip(index.keys, index.labels.tolist(), strict=True))
+
+    def counts(self) -> dict[str, int]:
+        """Return the number of references per state."""
+        index = self._index
+        if index is None:
+            return {}
+        states, counts = np.unique(index.labels, return_counts=True)
+        return dict(zip(states.tolist(), counts.tolist(), strict=True))
+
+    @property
+    def ready(self) -> bool:
+        """Return True once at least two states have references."""
+        return len(self.counts()) >= 2
+
+    def __len__(self) -> int:
+        """Return the number of references."""
+        index = self._index
+        return 0 if index is None else len(index.keys)
+
+    def classify(self, features: Features) -> Result:
+        """Classify ``features`` against the references."""
+        self._check(features.backend_id)
+        index = self._index
+        if index is None:
+            return Result(Status.NOT_READY)
+        states, counts = np.unique(index.labels, return_counts=True)
+        if len(states) < 2:
+            return Result(Status.NOT_READY)
+        similarity = (features.query @ index.matrix.T).max(axis=0)
+        k = min(self.k, int(counts.min()))
+        scores: dict[str, float] = {}
+        nearest: dict[str, str] = {}
+        for state in states.tolist():
+            members = np.flatnonzero(index.labels == state)
+            ranked = members[np.argsort(similarity[members])[::-1]]
+            scores[state] = float(similarity[ranked[:k]].mean())
+            nearest[state] = index.keys[ranked[0]]
+        best, runner_up = sorted(scores, key=scores.__getitem__, reverse=True)[:2]
+        margin = scores[best] - scores[runner_up]
+        confident = margin >= self.margin
+        return Result(
+            Status.CONFIDENT if confident else Status.UNSURE,
+            best if confident else None,
+            best,
+            margin,
+            scores,
+            nearest,
+        )
+
+    def _check(self, backend_id: str) -> None:
+        if backend_id != self.backend_id:
+            raise BackendMismatchError(
+                f"features from {backend_id}, expected {self.backend_id}"
+            )
+
+    def _build(self) -> _Index | None:
+        if not self._refs:
+            return None
+        keys = tuple(self._refs)
+        return _Index(
+            keys,
+            np.array([self._refs[key][0] for key in keys]),
+            np.stack([self._refs[key][1] for key in keys]),
+        )

tarsius/errors.py

@@ -0,0 +1,23 @@
+"""Exceptions raised by tarsius."""
+
+from __future__ import annotations
+
+
+class TarsiusError(Exception):
+    """Base class for all tarsius errors."""
+
+
+class ImageError(TarsiusError):
+    """The image could not be decoded, e.g. a camera returned an error page."""
+
+
+class BackendError(TarsiusError):
+    """A feature backend could not load its model or process an image."""
+
+
+class BackendMismatchError(TarsiusError, ValueError):
+    """Features come from another backend than the classifier expects."""
+
+
+class FeaturesError(TarsiusError):
+    """Stored features are corrupt or written by an incompatible version."""

tarsius/features.py

@@ -0,0 +1,101 @@
+"""Feature vectors and the backend protocol."""
+
+from __future__ import annotations
+
+import zipfile
+from dataclasses import dataclass
+from io import BytesIO
+from typing import Protocol
+
+import numpy as np
+import numpy.typing as npt
+from PIL import Image
+
+from .errors import FeaturesError
+from .roi import FULL_FRAME, Roi
+
+type Vector = npt.NDArray[np.float32]
+
+# Bump when the stored layout changes; older data then fails to load.
+_FORMAT = 1
+
+
+class FeatureBackend(Protocol):
+    """Turns an image region into vectors that compare by dot product."""
+
+    # Changes whenever extracted features change, so stored ones get rebuilt.
+    id: str
+    # Similarity gap between the two best states below which a result is unsure.
+    default_margin: float
+
+    def extract(self, image: Image.Image) -> tuple[Vector, Vector]:
+        """
+        Return ``(vector, query)`` for ``image``.
+
+        ``vector`` is the unit-norm reference form. ``query`` holds one or more
+        unit-norm rows; the similarity of a query to a reference is the best
+        row, which lets a backend tolerate small camera shifts.
+        """
+        ...
+
+
+# Arrays do not compare with ==, so identity equality.
+@dataclass(frozen=True, slots=True, eq=False)
+class Reference:
+    """The part of a confirmed frame a classifier keeps."""
+
+    backend_id: str
+    vector: Vector
+
+    def to_bytes(self) -> bytes:
+        """Serialize for a cache next to the original image."""
+        buffer = BytesIO()
+        np.savez(
+            buffer,
+            format=np.array(_FORMAT),
+            backend_id=np.array(self.backend_id),
+            vector=self.vector,
+        )
+        return buffer.getvalue()
+
+    @classmethod
+    def from_bytes(cls, data: bytes) -> Reference:
+        """Load data written by :meth:`to_bytes`; raise FeaturesError if unusable."""
+        try:
+            with np.load(BytesIO(data), allow_pickle=False) as npz:
+                if int(npz["format"]) != _FORMAT:
+                    raise FeaturesError(f"unsupported format {int(npz['format'])}")
+                return cls(str(npz["backend_id"]), npz["vector"].astype(np.float32))
+        except FeaturesError:
+            raise
+        except (EOFError, OSError, KeyError, ValueError, zipfile.BadZipFile) as err:
+            raise FeaturesError(f"unreadable features: {err}") from err
+
+
+@dataclass(frozen=True, slots=True, eq=False)
+class Features:
+    """Features of one image region."""
+
+    backend_id: str
+    vector: Vector
+    query: Vector
+
+    def reference(self) -> Reference:
+        """Return what a classifier keeps when this frame is confirmed."""
+        return Reference(self.backend_id, self.vector)
+
+
+def extract_features(
+    backend: FeatureBackend, image: Image.Image, roi: Roi = FULL_FRAME
+) -> Features:
+    """Crop ``image`` to ``roi`` and extract its features."""
+    vector, query = backend.extract(roi.crop(image))
+    return Features(backend.id, vector, query)
+
+
+def unit(array: npt.NDArray[np.floating]) -> Vector:
+    """Return ``array`` flattened, zero-mean and scaled to unit length."""
+    flat = array.astype(np.float32).ravel()
+    flat = flat - flat.mean()
+    norm = float(np.linalg.norm(flat))
+    return flat / norm if norm > 0 else flat

tarsius/image.py

@@ -0,0 +1,33 @@
+"""Image decoding."""
+
+from __future__ import annotations
+
+from io import BytesIO
+
+from PIL import Image, UnidentifiedImageError
+
+from .errors import ImageError
+
+
+def load_image(data: bytes, *, draft_side: int = 1024) -> Image.Image:
+    """
+    Decode an encoded image to RGB.
+
+    JPEGs are decoded at the smallest scale whose longer side is still at
+    least ``draft_side``, since full 4K decoding costs more than the feature
+    extraction itself. Other formats are decoded at full size.
+    """
+    if draft_side < 1:
+        raise ValueError("draft_side must be positive")
+    try:
+        image = Image.open(BytesIO(data))
+        width, height = image.size
+        scale = draft_side / max(width, height)
+        if scale < 1:
+            # draft keeps both sides at or above the request.
+            image.draft(
+                "RGB", (max(1, int(width * scale)), max(1, int(height * scale)))
+            )
+        return image.convert("RGB")
+    except (UnidentifiedImageError, OSError) as err:
+        raise ImageError(f"cannot decode image: {err}") from err

tarsius/models/LICENSE.squeezenet

@@ -0,0 +1,20 @@
+BSD LICENSE.
+
+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.
+
+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.

tarsius/models/squeezenet_v1.1.param

@@ -0,0 +1,50 @@
+7767517
+48 56
+Input                    data                     0 1 data -23330=5,3,227,227,1,3 0=227 1=227 2=3
+Convolution              conv1                    1 1 data conv1_relu_conv1 -23330=5,3,113,113,1,64 0=64 1=3 3=2 5=1 6=1728 9=1
+Pooling                  pool1                    1 1 conv1_relu_conv1 pool1 -23330=5,3,56,56,1,64 1=3 2=2
+Convolution              fire2/squeeze1x1         1 1 pool1 fire2/squeeze1x1_fire2/relu_squeeze1x1 -23330=5,3,56,56,1,16 0=16 1=1 5=1 6=1024 9=1
+Split                    splitncnn_0              1 2 fire2/squeeze1x1_fire2/relu_squeeze1x1 fire2/squeeze1x1_fire2/relu_squeeze1x1_splitncnn_0 fire2/squeeze1x1_fire2/relu_squeeze1x1_splitncnn_1 -23330=10,3,56,56,1,16,3,56,56,1,16
+Convolution              fire2/expand1x1          1 1 fire2/squeeze1x1_fire2/relu_squeeze1x1_splitncnn_1 fire2/expand1x1_fire2/relu_expand1x1 -23330=5,3,56,56,1,64 0=64 1=1 5=1 6=1024 9=1
+Convolution              fire2/expand3x3          1 1 fire2/squeeze1x1_fire2/relu_squeeze1x1_splitncnn_0 fire2/expand3x3_fire2/relu_expand3x3 -23330=5,3,56,56,1,64 0=64 1=3 4=1 5=1 6=9216 9=1
+Concat                   fire2/concat             2 1 fire2/expand1x1_fire2/relu_expand1x1 fire2/expand3x3_fire2/relu_expand3x3 fire2/concat -23330=5,3,56,56,1,128
+Convolution              fire3/squeeze1x1         1 1 fire2/concat fire3/squeeze1x1_fire3/relu_squeeze1x1 -23330=5,3,56,56,1,16 0=16 1=1 5=1 6=2048 9=1
+Split                    splitncnn_1              1 2 fire3/squeeze1x1_fire3/relu_squeeze1x1 fire3/squeeze1x1_fire3/relu_squeeze1x1_splitncnn_0 fire3/squeeze1x1_fire3/relu_squeeze1x1_splitncnn_1 -23330=10,3,56,56,1,16,3,56,56,1,16
+Convolution              fire3/expand1x1          1 1 fire3/squeeze1x1_fire3/relu_squeeze1x1_splitncnn_1 fire3/expand1x1_fire3/relu_expand1x1 -23330=5,3,56,56,1,64 0=64 1=1 5=1 6=1024 9=1
+Convolution              fire3/expand3x3          1 1 fire3/squeeze1x1_fire3/relu_squeeze1x1_splitncnn_0 fire3/expand3x3_fire3/relu_expand3x3 -23330=5,3,56,56,1,64 0=64 1=3 4=1 5=1 6=9216 9=1
+Concat                   fire3/concat             2 1 fire3/expand1x1_fire3/relu_expand1x1 fire3/expand3x3_fire3/relu_expand3x3 fire3/concat -23330=5,3,56,56,1,128
+Pooling                  pool3                    1 1 fire3/concat pool3 -23330=5,3,28,28,1,128 1=3 2=2
+Convolution              fire4/squeeze1x1         1 1 pool3 fire4/squeeze1x1_fire4/relu_squeeze1x1 -23330=5,3,28,28,1,32 0=32 1=1 5=1 6=4096 9=1
+Split                    splitncnn_2              1 2 fire4/squeeze1x1_fire4/relu_squeeze1x1 fire4/squeeze1x1_fire4/relu_squeeze1x1_splitncnn_0 fire4/squeeze1x1_fire4/relu_squeeze1x1_splitncnn_1 -23330=10,3,28,28,1,32,3,28,28,1,32
+Convolution              fire4/expand1x1          1 1 fire4/squeeze1x1_fire4/relu_squeeze1x1_splitncnn_1 fire4/expand1x1_fire4/relu_expand1x1 -23330=5,3,28,28,1,128 0=128 1=1 5=1 6=4096 9=1
+Convolution              fire4/expand3x3          1 1 fire4/squeeze1x1_fire4/relu_squeeze1x1_splitncnn_0 fire4/expand3x3_fire4/relu_expand3x3 -23330=5,3,28,28,1,128 0=128 1=3 4=1 5=1 6=36864 9=1
+Concat                   fire4/concat             2 1 fire4/expand1x1_fire4/relu_expand1x1 fire4/expand3x3_fire4/relu_expand3x3 fire4/concat -23330=5,3,28,28,1,256
+Convolution              fire5/squeeze1x1         1 1 fire4/concat fire5/squeeze1x1_fire5/relu_squeeze1x1 -23330=5,3,28,28,1,32 0=32 1=1 5=1 6=8192 9=1
+Split                    splitncnn_3              1 2 fire5/squeeze1x1_fire5/relu_squeeze1x1 fire5/squeeze1x1_fire5/relu_squeeze1x1_splitncnn_0 fire5/squeeze1x1_fire5/relu_squeeze1x1_splitncnn_1 -23330=10,3,28,28,1,32,3,28,28,1,32
+Convolution              fire5/expand1x1          1 1 fire5/squeeze1x1_fire5/relu_squeeze1x1_splitncnn_1 fire5/expand1x1_fire5/relu_expand1x1 -23330=5,3,28,28,1,128 0=128 1=1 5=1 6=4096 9=1
+Convolution              fire5/expand3x3          1 1 fire5/squeeze1x1_fire5/relu_squeeze1x1_splitncnn_0 fire5/expand3x3_fire5/relu_expand3x3 -23330=5,3,28,28,1,128 0=128 1=3 4=1 5=1 6=36864 9=1
+Concat                   fire5/concat             2 1 fire5/expand1x1_fire5/relu_expand1x1 fire5/expand3x3_fire5/relu_expand3x3 fire5/concat -23330=5,3,28,28,1,256
+Pooling                  pool5                    1 1 fire5/concat pool5 -23330=5,3,14,14,1,256 1=3 2=2
+Convolution              fire6/squeeze1x1         1 1 pool5 fire6/squeeze1x1_fire6/relu_squeeze1x1 -23330=5,3,14,14,1,48 0=48 1=1 5=1 6=12288 9=1
+Split                    splitncnn_4              1 2 fire6/squeeze1x1_fire6/relu_squeeze1x1 fire6/squeeze1x1_fire6/relu_squeeze1x1_splitncnn_0 fire6/squeeze1x1_fire6/relu_squeeze1x1_splitncnn_1 -23330=10,3,14,14,1,48,3,14,14,1,48
+Convolution              fire6/expand1x1          1 1 fire6/squeeze1x1_fire6/relu_squeeze1x1_splitncnn_1 fire6/expand1x1_fire6/relu_expand1x1 -23330=5,3,14,14,1,192 0=192 1=1 5=1 6=9216 9=1
+Convolution              fire6/expand3x3          1 1 fire6/squeeze1x1_fire6/relu_squeeze1x1_splitncnn_0 fire6/expand3x3_fire6/relu_expand3x3 -23330=5,3,14,14,1,192 0=192 1=3 4=1 5=1 6=82944 9=1
+Concat                   fire6/concat             2 1 fire6/expand1x1_fire6/relu_expand1x1 fire6/expand3x3_fire6/relu_expand3x3 fire6/concat -23330=5,3,14,14,1,384
+Convolution              fire7/squeeze1x1         1 1 fire6/concat fire7/squeeze1x1_fire7/relu_squeeze1x1 -23330=5,3,14,14,1,48 0=48 1=1 5=1 6=18432 9=1
+Split                    splitncnn_5              1 2 fire7/squeeze1x1_fire7/relu_squeeze1x1 fire7/squeeze1x1_fire7/relu_squeeze1x1_splitncnn_0 fire7/squeeze1x1_fire7/relu_squeeze1x1_splitncnn_1 -23330=10,3,14,14,1,48,3,14,14,1,48
+Convolution              fire7/expand1x1          1 1 fire7/squeeze1x1_fire7/relu_squeeze1x1_splitncnn_1 fire7/expand1x1_fire7/relu_expand1x1 -23330=5,3,14,14,1,192 0=192 1=1 5=1 6=9216 9=1
+Convolution              fire7/expand3x3          1 1 fire7/squeeze1x1_fire7/relu_squeeze1x1_splitncnn_0 fire7/expand3x3_fire7/relu_expand3x3 -23330=5,3,14,14,1,192 0=192 1=3 4=1 5=1 6=82944 9=1
+Concat                   fire7/concat             2 1 fire7/expand1x1_fire7/relu_expand1x1 fire7/expand3x3_fire7/relu_expand3x3 fire7/concat -23330=5,3,14,14,1,384
+Convolution              fire8/squeeze1x1         1 1 fire7/concat fire8/squeeze1x1_fire8/relu_squeeze1x1 -23330=5,3,14,14,1,64 0=64 1=1 5=1 6=24576 9=1
+Split                    splitncnn_6              1 2 fire8/squeeze1x1_fire8/relu_squeeze1x1 fire8/squeeze1x1_fire8/relu_squeeze1x1_splitncnn_0 fire8/squeeze1x1_fire8/relu_squeeze1x1_splitncnn_1 -23330=10,3,14,14,1,64,3,14,14,1,64
+Convolution              fire8/expand1x1          1 1 fire8/squeeze1x1_fire8/relu_squeeze1x1_splitncnn_1 fire8/expand1x1_fire8/relu_expand1x1 -23330=5,3,14,14,1,256 0=256 1=1 5=1 6=16384 9=1
+Convolution              fire8/expand3x3          1 1 fire8/squeeze1x1_fire8/relu_squeeze1x1_splitncnn_0 fire8/expand3x3_fire8/relu_expand3x3 -23330=5,3,14,14,1,256 0=256 1=3 4=1 5=1 6=147456 9=1
+Concat                   fire8/concat             2 1 fire8/expand1x1_fire8/relu_expand1x1 fire8/expand3x3_fire8/relu_expand3x3 fire8/concat -23330=5,3,14,14,1,512
+Convolution              fire9/squeeze1x1         1 1 fire8/concat fire9/squeeze1x1_fire9/relu_squeeze1x1 -23330=5,3,14,14,1,64 0=64 1=1 5=1 6=32768 9=1
+Split                    splitncnn_7              1 2 fire9/squeeze1x1_fire9/relu_squeeze1x1 fire9/squeeze1x1_fire9/relu_squeeze1x1_splitncnn_0 fire9/squeeze1x1_fire9/relu_squeeze1x1_splitncnn_1 -23330=10,3,14,14,1,64,3,14,14,1,64
+Convolution              fire9/expand1x1          1 1 fire9/squeeze1x1_fire9/relu_squeeze1x1_splitncnn_1 fire9/expand1x1_fire9/relu_expand1x1 -23330=5,3,14,14,1,256 0=256 1=1 5=1 6=16384 9=1
+Convolution              fire9/expand3x3          1 1 fire9/squeeze1x1_fire9/relu_squeeze1x1_splitncnn_0 fire9/expand3x3_fire9/relu_expand3x3 -23330=5,3,14,14,1,256 0=256 1=3 4=1 5=1 6=147456 9=1
+Concat                   fire9/concat             2 1 fire9/expand1x1_fire9/relu_expand1x1 fire9/expand3x3_fire9/relu_expand3x3 fire9/concat_drop9 -23330=5,3,14,14,1,512
+Convolution              conv10                   1 1 fire9/concat_drop9 conv10_relu_conv10 -23330=5,3,14,14,1,1000 0=1000 1=1 5=1 6=512000 9=1
+Pooling                  pool10                   1 1 conv10_relu_conv10 pool10 -23330=5,1,1000,1,1,1 0=1 4=1
+Softmax                  prob                     1 1 pool10 prob -23330=5,1,1000,1,1,1

tarsius/roi.py

@@ -0,0 +1,59 @@
+"""Region of interest inside a camera image."""
+
+from __future__ import annotations
+
+from collections.abc import Iterable
+from dataclasses import astuple, dataclass
+
+from PIL import Image
+
+
+@dataclass(frozen=True, slots=True)
+class Roi:
+    """Rectangle as fractions of the image size, so it survives resolution changes."""
+
+    left: float = 0.0
+    top: float = 0.0
+    right: float = 1.0
+    bottom: float = 1.0
+
+    def __post_init__(self) -> None:
+        """Reject empty or out-of-frame rectangles."""
+        if not (0.0 <= self.left < self.right <= 1.0):
+            raise ValueError(f"invalid horizontal range {self.left}..{self.right}")
+        if not (0.0 <= self.top < self.bottom <= 1.0):
+            raise ValueError(f"invalid vertical range {self.top}..{self.bottom}")
+
+    @classmethod
+    def from_cells(
+        cls, cells: Iterable[tuple[int, int]], *, columns: int, rows: int
+    ) -> Roi:
+        """Return the bounding box of grid cells given as ``(column, row)``."""
+        picked = list(cells)
+        if not picked:
+            raise ValueError("no cells selected")
+        for column, row in picked:
+            if not (0 <= column < columns and 0 <= row < rows):
+                raise ValueError(f"cell {column},{row} outside a {columns}x{rows} grid")
+        return cls(
+            min(column for column, _ in picked) / columns,
+            min(row for _, row in picked) / rows,
+            (max(column for column, _ in picked) + 1) / columns,
+            (max(row for _, row in picked) + 1) / rows,
+        )
+
+    def as_tuple(self) -> tuple[float, float, float, float]:
+        """Return ``(left, top, right, bottom)``, e.g. for config storage."""
+        return astuple(self)
+
+    def crop(self, image: Image.Image) -> Image.Image:
+        """Return the region of ``image``, at least one pixel in each direction."""
+        width, height = image.size
+        left = min(round(self.left * width), width - 1)
+        top = min(round(self.top * height), height - 1)
+        right = max(round(self.right * width), left + 1)
+        bottom = max(round(self.bottom * height), top + 1)
+        return image.crop((left, top, right, bottom))
+
+
+FULL_FRAME = Roi()

tarsius/vote.py

@@ -0,0 +1,27 @@
+"""Combine several frames into one decision."""
+
+from __future__ import annotations
+
+from collections import Counter
+from collections.abc import Sequence
+
+from .classifier import Result
+
+
+def majority(results: Sequence[Result], *, min_share: float = 0.5) -> str | None:
+    """
+    Return the state picked by more than ``min_share`` of all frames.
+
+    Unsure and not-ready frames count against every state, so a burst of
+    unsure frames keeps the decision open instead of letting one confident
+    frame win. ``min_share`` of at least 0.5 makes the winner unique.
+    """
+    if not 0.5 <= min_share < 1:
+        raise ValueError("min_share must be in [0.5, 1)")
+    if not results:
+        return None
+    votes = Counter(result.state for result in results if result.state is not None)
+    if not votes:
+        return None
+    state, count = votes.most_common(1)[0]
+    return state if count / len(results) > min_share else None

tarsius-0.1.0.dist-info/METADATA

@@ -0,0 +1,106 @@
+Metadata-Version: 2.5
+Name: tarsius
+Version: 0.1.0
+Summary: Local, training-free state detection for a region of a camera image
+Project-URL: Repository, https://github.com/RaHehl/tarsius
+Project-URL: Bug Tracker, https://github.com/RaHehl/tarsius/issues
+Project-URL: Changelog, https://github.com/RaHehl/tarsius/releases
+Author: Raphael Hehl
+License-Expression: Apache-2.0 AND BSD-2-Clause
+License-File: LICENSE
+License-File: NOTICE
+License-File: src/tarsius/models/LICENSE.squeezenet
+Keywords: camera,computer-vision,home-assistant,ncnn,presence
+Classifier: Development Status :: 2 - Pre-Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering :: Image Recognition
+Classifier: Topic :: Software Development :: Libraries
+Classifier: Typing :: Typed
+Requires-Python: >=3.14
+Requires-Dist: numpy>=2.0
+Requires-Dist: pillow>=11.0
+Provides-Extra: ncnn
+Requires-Dist: ncnn>=1.0.20260526; extra == 'ncnn'
+Description-Content-Type: text/markdown
+
+# tarsius
+
+<p align="center">
+  <a href="https://github.com/RaHehl/tarsius/actions/workflows/ci.yml?query=branch%3Amain"><img src="https://github.com/RaHehl/tarsius/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
+  <a href="https://codecov.io/gh/RaHehl/tarsius"><img src="https://codecov.io/gh/RaHehl/tarsius/graph/badge.svg" alt="Coverage"></a>
+  <a href="https://pypi.org/project/tarsius/"><img src="https://img.shields.io/pypi/v/tarsius" alt="PyPI"></a>
+  <a href="https://pypi.org/project/tarsius/"><img src="https://img.shields.io/pypi/pyversions/tarsius" alt="Python"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/license-Apache--2.0-blue" alt="License"></a>
+</p>
+<p align="center">
+  <a href="https://github.com/astral-sh/uv"><img src="https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/uv/main/assets/badge/v0.json" alt="uv"></a>
+  <a href="https://github.com/astral-sh/ruff"><img src="https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json" alt="Ruff"></a>
+  <a href="https://github.com/j178/prek"><img src="https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/j178/prek/master/docs/assets/badge-v0.json" alt="prek"></a>
+</p>
+
+Local, training-free state detection for a region of a camera image: is the car in the garage, is the mailbox flag up, is the gate open.
+
+You confirm a few frames per state, tarsius compares new frames against them and answers with a state or "unsure". Nothing leaves the machine and nothing gets trained.
+
+```python
+from tarsius import Classifier, NcnnBackend, Roi, extract_features, load_image
+
+backend = NcnnBackend()
+spot = Roi(0.10, 0.10, 0.60, 0.50)
+classifier = Classifier.for_backend(backend)
+car = extract_features(backend, load_image(car_jpeg), spot)
+empty = extract_features(backend, load_image(empty_jpeg), spot)
+classifier.add("ref-1", car, state="occupied")
+classifier.add("ref-2", empty, state="free")
+
+result = classifier.classify(extract_features(backend, load_image(new_jpeg), spot))
+result.status  # Status.CONFIDENT, Status.UNSURE or Status.NOT_READY
+result.state  # "occupied" or "free" when confident
+result.candidate  # best guess, also when unsure
+```
+
+Store `features.reference().to_bytes()` next to each confirmed frame and restore it with `Reference.from_bytes()`. On `FeaturesError`, extract again from the stored frame.
+
+## Backends
+
+- `NcnnBackend`: SqueezeNet 1.1 ImageNet features on [ncnn](https://github.com/Tencent/ncnn), install with `tarsius[ncnn]`. Tolerates clutter and lighting changes, a rough region is enough.
+- `EdgeBackend`: numpy only, a baseline. Needs a tight region that still shows some static structure when the object is gone.
+
+`NcnnBackend` is the intended default. It stays an optional extra until ncnn ships a release that no longer requires `opencv-python`, which has no musllinux wheels. Without that, tarsius cannot install on Alpine (musl) systems such as Home Assistant OS, whose Core container is Alpine based. ncnn master already made opencv optional ([Tencent/ncnn@0ca5a4b](https://github.com/Tencent/ncnn/commit/0ca5a4b6d)), so the next ncnn release should unblock it.
+
+## Results so far
+
+One household, two garage cameras, 24 days of recorded history, day and IR night.
+References come from older frames, the test is the newest week.
+The table is the harder camera: a dark car and clutter that gets moved around in front of it (459 frames, 164 in the test week).
+The other camera scores 100 % with both backends.
+
+| Backend | Region | 2 confirmed segments per state | 5 per state |
+|---|---|---|---|
+| ncnn | tight | 0 % wrong, 0 % unsure | 0 % wrong, 0 % unsure |
+| ncnn | wide, with clutter | 0 % wrong, 0 % unsure | 0 % wrong, 0 % unsure |
+| edges | tight | 1.8 % wrong, 3.5 % unsure | 0 % wrong, 1.8 % unsure |
+| edges | wide, with clutter | 27 % wrong, 7.6 % unsure | 20 % wrong, 8.5 % unsure |
+
+About 20 ms per 4 MP frame including JPEG decoding, for either backend, on an x86 desktop with one thread.
+A small sample, not a guarantee. `scripts/history_split.py` runs the same evaluation on your own recordings.
+
+## Development
+
+```sh
+scripts/build_model.sh   # once, builds the ncnn model into src/tarsius/models
+uv sync --extra ncnn
+uv run pytest
+uvx prek run --all-files
+```
+
+The ncnn model is not stored in git. `scripts/build_model.sh` converts the original SqueezeNet 1.1 release with the ncnn tools. Inputs and outputs are checked against pinned sha256 sums, so CI and a local run produce the same files. It needs x86_64 Linux with `curl`, `unzip`, `libgomp1` and `libprotobuf32t64` (Ubuntu 24.04).
+
+Releases: every PR gets a label, Release Drafter collects them into a draft, publishing the GitHub release builds and uploads to PyPI.
+
+## License
+
+Apache-2.0. The bundled SqueezeNet model is BSD-2-Clause, see NOTICE.

tarsius-0.1.0.dist-info/RECORD

@@ -0,0 +1,20 @@
+tarsius/__init__.py,sha256=FGS7mfLH-C9Q04uAye2lSUOBrmWfqvPE9rmdC8OO4rI,915
+tarsius/classifier.py,sha256=d9mDZ_SOLuXkZcHCIad3z_tFyYGZO9qrHWalzz2-mOo,6409
+tarsius/errors.py,sha256=1VLJa1zl7wgLjuyXbrq1LG0zF2EtR_GsqTckCVl9M90,618
+tarsius/features.py,sha256=SDC3RSXaC-QS-fl7JfwNQUxqYpkisq5VKOdeetNCtJc,3224
+tarsius/image.py,sha256=hPj6GLT6LJc775vVHqDCGebVf25tjX8SI3achhZTQ30,1072
+tarsius/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+tarsius/roi.py,sha256=v6ZX7_Byvu9Eb0YhgcpqX_VJmYnGCxH1ZR6-G4nN41o,2187
+tarsius/vote.py,sha256=MrjKVlvBUz8IqDdd6k43B7dU9wvYJbqWRirFOYruofk,934
+tarsius/backends/__init__.py,sha256=UrpCqWYPwhcD422uFmq11Bxw-X1uX5x1pIn8HOJi6kc,24
+tarsius/backends/edges.py,sha256=bHHd1ZEa_kcJ-r1nf0aoqANDXSy0JB1kkEJNhE8xVvk,1698
+tarsius/backends/ncnn.py,sha256=4zOTl8yTBklatbM1zB2bvb3FGw666zDr8DmLGbq9tNc,3112
+tarsius/models/LICENSE.squeezenet,sha256=kA4XbjG3IKi7Py2xznNfSaLCjtzO4LsfvtKzUMMYtdA,1250
+tarsius/models/squeezenet_v1.1.bin,sha256=upJbC7ya6uo0uqKlgFaZDakrpAYlAgm84DgvXGbicls,2478984
+tarsius/models/squeezenet_v1.1.param,sha256=ylWFUgqxKBpE-ifrBumFGj3-iCsheXjWza3d1YNWR3w,8336
+tarsius-0.1.0.dist-info/METADATA,sha256=lFq1K8erf28uZBh-ydnJCIJ86vj13M5kAyilUq5T0ac,5997
+tarsius-0.1.0.dist-info/WHEEL,sha256=W3fkpkm7-wf9vBI5Z-7s0eWkeM-spu78I8Neb98DeEg,87
+tarsius-0.1.0.dist-info/licenses/LICENSE,sha256=z8d0m5b2O9McPEK1xHG_dWgUBT6EfBDz6wA0F7xSPTA,11358
+tarsius-0.1.0.dist-info/licenses/NOTICE,sha256=lxrGjKZXPCPfUiINzVR3cWK-1ibZnulgoMRCLSRk0S0,332
+tarsius-0.1.0.dist-info/licenses/src/tarsius/models/LICENSE.squeezenet,sha256=kA4XbjG3IKi7Py2xznNfSaLCjtzO4LsfvtKzUMMYtdA,1250
+tarsius-0.1.0.dist-info/RECORD,,

tarsius-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.32.4
+Root-Is-Purelib: true
+Tag: py3-none-any

tarsius-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

tarsius-0.1.0.dist-info/licenses/NOTICE

@@ -0,0 +1,8 @@
+tarsius
+Copyright 2026 Raphael Hehl
+
+This product bundles the SqueezeNet v1.1 model
+(https://github.com/forresti/SqueezeNet), converted to ncnn with fp16
+weights. BSD-2-Clause, see tarsius/models/LICENSE.squeezenet.
+SqueezeNet by Forrest N. Iandola, Song Han, Matthew W. Moskewicz,
+Khalid Ashraf, William J. Dally and Kurt Keutzer.

tarsius-0.1.0.dist-info/licenses/src/tarsius/models/LICENSE.squeezenet

@@ -0,0 +1,20 @@
+BSD LICENSE.
+
+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.
+
+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.