python-doctr 0.8.0__py3-none-any.whl → 0.9.0__py3-none-any.whl

doctr/models/artefacts/barcode.py

@@ -1,74 +0,0 @@
-# Copyright (C) 2021-2024, Mindee.
-
-# This program is licensed under the Apache License 2.0.
-# See LICENSE or go to <https://opensource.org/licenses/Apache-2.0> for full license details.
-
-from typing import List, Tuple
-
-import cv2
-import numpy as np
-
-__all__ = ["BarCodeDetector"]
-
-
-class BarCodeDetector:
-    """Implements a Bar-code detector.
-    For now, only horizontal (or with a small angle) bar-codes are supported
-
-    Args:
-    ----
-        min_size: minimum relative size of a barcode on the page
-        canny_minval: lower bound for canny hysteresis
-        canny_maxval: upper-bound for canny hysteresis
-    """
-
-    def __init__(self, min_size: float = 1 / 6, canny_minval: int = 50, canny_maxval: int = 150) -> None:
-        self.min_size = min_size
-        self.canny_minval = canny_minval
-        self.canny_maxval = canny_maxval
-
-    def __call__(
-        self,
-        img: np.ndarray,
-    ) -> List[Tuple[float, float, float, float]]:
-        """Detect Barcodes on the image
-        Args:
-            img: np image
-
-        Returns
-        -------
-            A list of tuples: [(xmin, ymin, xmax, ymax), ...] containing barcodes rel. coordinates
-        """
-        # get image size and define parameters
-        height, width = img.shape[:2]
-        k = (1 + int(width / 512)) * 10  # spatial extension of kernels, 512 -> 20, 1024 -> 30, ...
-        min_w = int(width * self.min_size)  # minimal size of a possible barcode
-
-        # Detect edges
-        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-        edges = cv2.Canny(gray, self.canny_minval, self.canny_maxval, apertureSize=3)
-
-        # Horizontal dilation to aggregate bars of the potential barcode
-        # without aggregating text lines of the page vertically
-        edges = cv2.dilate(edges, np.ones((1, k), np.uint8))
-
-        # Instantiate a barcode-shaped kernel and erode to keep only vertical-bar structures
-        bar_code_kernel: np.ndarray = np.zeros((k, 3), np.uint8)
-        bar_code_kernel[..., [0, 2]] = 1
-        edges = cv2.erode(edges, bar_code_kernel, iterations=1)
-
-        # Opening to remove noise
-        edges = cv2.morphologyEx(edges, cv2.MORPH_OPEN, np.ones((k, k), np.uint8))
-
-        # Dilation to retrieve vertical length (lost at the first dilation)
-        edges = cv2.dilate(edges, np.ones((k, 1), np.uint8))
-
-        # Find contours, and keep the widest as barcodes
-        contours, _ = cv2.findContours(edges, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
-        barcodes = []
-        for contour in contours:
-            x, y, w, h = cv2.boundingRect(contour)
-            if w >= min_w:
-                barcodes.append((x / width, y / height, (x + w) / width, (y + h) / height))
-
-        return barcodes

doctr/models/artefacts/face.py

@@ -1,63 +0,0 @@
-# Copyright (C) 2021-2024, Mindee.
-
-# This program is licensed under the Apache License 2.0.
-# See LICENSE or go to <https://opensource.org/licenses/Apache-2.0> for full license details.
-
-from typing import List, Tuple
-
-import cv2
-import numpy as np
-
-from doctr.utils.repr import NestedObject
-
-__all__ = ["FaceDetector"]
-
-
-class FaceDetector(NestedObject):
-    """Implements a face detector to detect profile pictures on resumes, IDS, driving licenses, passports...
-    Based on open CV CascadeClassifier (haarcascades)
-
-    Args:
-    ----
-        n_faces: maximal number of faces to detect on a single image, default = 1
-    """
-
-    def __init__(
-        self,
-        n_faces: int = 1,
-    ) -> None:
-        self.n_faces = n_faces
-        # Instantiate classifier
-        self.detector = cv2.CascadeClassifier(
-            cv2.data.haarcascades + "haarcascade_frontalface_default.xml"  # type: ignore[attr-defined]
-        )
-
-    def extra_repr(self) -> str:
-        return f"n_faces={self.n_faces}"
-
-    def __call__(
-        self,
-        img: np.ndarray,
-    ) -> List[Tuple[float, float, float, float]]:
-        """Detect n_faces on the img
-
-        Args:
-        ----
-            img: image to detect faces on
-
-        Returns:
-        -------
-            A list of size n_faces, each face is a tuple of relative xmin, ymin, xmax, ymax
-        """
-        height, width = img.shape[:2]
-        gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
-
-        faces = self.detector.detectMultiScale(gray, 1.5, 3)
-        # If faces are detected, keep only the biggest ones
-        rel_faces = []
-        if len(faces) > 0:
-            x, y, w, h = sorted(faces, key=lambda x: x[2] + x[3])[-min(self.n_faces, len(faces))]
-            xmin, ymin, xmax, ymax = x / width, y / height, (x + w) / width, (y + h) / height
-            rel_faces.append((xmin, ymin, xmax, ymax))
-
-        return rel_faces

doctr/models/obj_detection/__init__.py

@@ -1 +0,0 @@
-from .faster_rcnn import *

doctr/models/obj_detection/faster_rcnn/__init__.py

@@ -1,4 +0,0 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if not is_tf_available() and is_torch_available():
-    from .pytorch import *

doctr/models/obj_detection/faster_rcnn/pytorch.py

@@ -1,81 +0,0 @@
-# Copyright (C) 2021-2024, Mindee.
-
-# This program is licensed under the Apache License 2.0.
-# See LICENSE or go to <https://opensource.org/licenses/Apache-2.0> for full license details.
-
-from typing import Any, Dict
-
-from torchvision.models.detection import FasterRCNN, FasterRCNN_MobileNet_V3_Large_FPN_Weights, faster_rcnn
-
-from ...utils import load_pretrained_params
-
-__all__ = ["fasterrcnn_mobilenet_v3_large_fpn"]
-
-
-default_cfgs: Dict[str, Dict[str, Any]] = {
-    "fasterrcnn_mobilenet_v3_large_fpn": {
-        "input_shape": (3, 1024, 1024),
-        "mean": (0.485, 0.456, 0.406),
-        "std": (0.229, 0.224, 0.225),
-        "classes": ["background", "qr_code", "bar_code", "logo", "photo"],
-        "url": "https://doctr-static.mindee.com/models?id=v0.4.1/fasterrcnn_mobilenet_v3_large_fpn-d5b2490d.pt&src=0",
-    },
-}
-
-
-def _fasterrcnn(arch: str, pretrained: bool, **kwargs: Any) -> FasterRCNN:
-    _kwargs = {
-        "image_mean": default_cfgs[arch]["mean"],
-        "image_std": default_cfgs[arch]["std"],
-        "box_detections_per_img": 150,
-        "box_score_thresh": 0.5,
-        "box_positive_fraction": 0.35,
-        "box_nms_thresh": 0.2,
-        "rpn_nms_thresh": 0.2,
-        "num_classes": len(default_cfgs[arch]["classes"]),
-    }
-
-    # Build the model
-    _kwargs.update(kwargs)
-    model = faster_rcnn.__dict__[arch](weights=None, weights_backbone=None, **_kwargs)
-    model.cfg = default_cfgs[arch]
-
-    if pretrained:
-        # Load pretrained parameters
-        load_pretrained_params(model, default_cfgs[arch]["url"])
-    else:
-        # Filter keys
-        state_dict = {
-            k: v
-            for k, v in faster_rcnn.__dict__[arch](weights=FasterRCNN_MobileNet_V3_Large_FPN_Weights.DEFAULT)
-            .state_dict()
-            .items()
-            if not k.startswith("roi_heads.")
-        }
-
-        # Load state dict
-        model.load_state_dict(state_dict, strict=False)
-
-    return model
-
-
-def fasterrcnn_mobilenet_v3_large_fpn(pretrained: bool = False, **kwargs: Any) -> FasterRCNN:
-    """Faster-RCNN architecture with a MobileNet V3 backbone as described in `"Faster R-CNN: Towards Real-Time
-    Object Detection with Region Proposal Networks" <https://arxiv.org/pdf/1506.01497.pdf>`_.
-
-    >>> import torch
-    >>> from doctr.models.obj_detection import fasterrcnn_mobilenet_v3_large_fpn
-    >>> model = fasterrcnn_mobilenet_v3_large_fpn(pretrained=True)
-    >>> input_tensor = torch.rand((1, 3, 1024, 1024), dtype=torch.float32)
-    >>> out = model(input_tensor)
-
-    Args:
-    ----
-        pretrained (bool): If True, returns a model pre-trained on our object detection dataset
-        **kwargs: keyword arguments of the FasterRCNN architecture
-
-    Returns:
-    -------
-        object detection architecture
-    """
-    return _fasterrcnn("fasterrcnn_mobilenet_v3_large_fpn", pretrained, **kwargs)