python-doctr 0.9.0__py3-none-any.whl → 0.11.0__py3-none-any.whl

doctr/datasets/sroie.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
@@ -6,7 +6,7 @@
 import csv
 import os
 from pathlib import Path
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import numpy as np
 from tqdm import tqdm
@@ -29,10 +29,10 @@ class SROIE(VisionDataset):
     >>> img, target = train_set[0]
 
     Args:
-    ----
         train: whether the subset should be the training one
         use_polygons: whether polygons should be considered as rotated bounding box (instead of straight ones)
         recognition_task: whether the dataset should be used for recognition task
+        detection_task: whether the dataset should be used for detection task
         **kwargs: keyword arguments from `VisionDataset`.
     """
 
@@ -52,6 +52,7 @@ class SROIE(VisionDataset):
         train: bool = True,
         use_polygons: bool = False,
         recognition_task: bool = False,
+        detection_task: bool = False,
         **kwargs: Any,
     ) -> None:
         url, sha256, name = self.TRAIN if train else self.TEST
@@ -63,13 +64,21 @@ class SROIE(VisionDataset):
             pre_transforms=convert_target_to_relative if not recognition_task else None,
             **kwargs,
         )
+        if recognition_task and detection_task:
+            raise ValueError(
+                "`recognition_task` and `detection_task` cannot be set to True simultaneously. "
+                + "To get the whole dataset with boxes and labels leave both parameters to False."
+            )
+
         self.train = train
 
         tmp_root = os.path.join(self.root, "images")
-        self.data: List[Tuple[Union[str, np.ndarray], Union[str, Dict[str, Any]]]] = []
+        self.data: list[tuple[str | np.ndarray, str | dict[str, Any] | np.ndarray]] = []
         np_dtype = np.float32
 
-        for img_path in tqdm(iterable=os.listdir(tmp_root), desc="Unpacking SROIE", total=len(os.listdir(tmp_root))):
+        for img_path in tqdm(
+            iterable=os.listdir(tmp_root), desc="Preparing and Loading SROIE", total=len(os.listdir(tmp_root))
+        ):
             # File existence check
             if not os.path.exists(os.path.join(tmp_root, img_path)):
                 raise FileNotFoundError(f"unable to locate {os.path.join(tmp_root, img_path)}")
@@ -94,6 +103,8 @@ class SROIE(VisionDataset):
                 for crop, label in zip(crops, labels):
                     if crop.shape[0] > 0 and crop.shape[1] > 0 and len(label) > 0:
                         self.data.append((crop, label))
+            elif detection_task:
+                self.data.append((img_path, coords))
             else:
                 self.data.append((img_path, dict(boxes=coords, labels=labels)))
 

doctr/datasets/svhn.py

@@ -1,10 +1,10 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
 
 import os
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import h5py
 import numpy as np
@@ -28,10 +28,10 @@ class SVHN(VisionDataset):
     >>> img, target = train_set[0]
 
     Args:
-    ----
         train: whether the subset should be the training one
         use_polygons: whether polygons should be considered as rotated bounding box (instead of straight ones)
         recognition_task: whether the dataset should be used for recognition task
+        detection_task: whether the dataset should be used for detection task
         **kwargs: keyword arguments from `VisionDataset`.
     """
 
@@ -52,6 +52,7 @@ class SVHN(VisionDataset):
         train: bool = True,
         use_polygons: bool = False,
         recognition_task: bool = False,
+        detection_task: bool = False,
         **kwargs: Any,
     ) -> None:
         url, sha256, name = self.TRAIN if train else self.TEST
@@ -63,8 +64,14 @@ class SVHN(VisionDataset):
             pre_transforms=convert_target_to_relative if not recognition_task else None,
             **kwargs,
         )
+        if recognition_task and detection_task:
+            raise ValueError(
+                "`recognition_task` and `detection_task` cannot be set to True simultaneously. "
+                + "To get the whole dataset with boxes and labels leave both parameters to False."
+            )
+
         self.train = train
-        self.data: List[Tuple[Union[str, np.ndarray], Union[str, Dict[str, Any]]]] = []
+        self.data: list[tuple[str | np.ndarray, str | dict[str, Any] | np.ndarray]] = []
         np_dtype = np.float32
 
         tmp_root = os.path.join(self.root, "train" if train else "test")
@@ -73,7 +80,9 @@ class SVHN(VisionDataset):
         with h5py.File(os.path.join(tmp_root, "digitStruct.mat"), "r") as f:
             img_refs = f["digitStruct/name"]
             box_refs = f["digitStruct/bbox"]
-            for img_ref, box_ref in tqdm(iterable=zip(img_refs, box_refs), desc="Unpacking SVHN", total=len(img_refs)):
+            for img_ref, box_ref in tqdm(
+                iterable=zip(img_refs, box_refs), desc="Preparing and Loading SVHN", total=len(img_refs)
+            ):
                 # convert ascii matrix to string
                 img_name = "".join(map(chr, f[img_ref[0]][()].flatten()))
 
@@ -122,6 +131,8 @@ class SVHN(VisionDataset):
                     for crop, label in zip(crops, label_targets):
                         if crop.shape[0] > 0 and crop.shape[1] > 0 and len(label) > 0:
                             self.data.append((crop, label))
+                elif detection_task:
+                    self.data.append((img_name, box_targets))
                 else:
                     self.data.append((img_name, dict(boxes=box_targets, labels=label_targets)))
 

doctr/datasets/svt.py

@@ -1,10 +1,10 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
 
 import os
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import defusedxml.ElementTree as ET
 import numpy as np
@@ -28,10 +28,10 @@ class SVT(VisionDataset):
     >>> img, target = train_set[0]
 
     Args:
-    ----
         train: whether the subset should be the training one
         use_polygons: whether polygons should be considered as rotated bounding box (instead of straight ones)
         recognition_task: whether the dataset should be used for recognition task
+        detection_task: whether the dataset should be used for detection task
         **kwargs: keyword arguments from `VisionDataset`.
     """
 
@@ -43,6 +43,7 @@ class SVT(VisionDataset):
         train: bool = True,
         use_polygons: bool = False,
         recognition_task: bool = False,
+        detection_task: bool = False,
         **kwargs: Any,
     ) -> None:
         super().__init__(
@@ -53,8 +54,14 @@ class SVT(VisionDataset):
             pre_transforms=convert_target_to_relative if not recognition_task else None,
             **kwargs,
         )
+        if recognition_task and detection_task:
+            raise ValueError(
+                "`recognition_task` and `detection_task` cannot be set to True simultaneously. "
+                + "To get the whole dataset with boxes and labels leave both parameters to False."
+            )
+
         self.train = train
-        self.data: List[Tuple[Union[str, np.ndarray], Union[str, Dict[str, Any]]]] = []
+        self.data: list[tuple[str | np.ndarray, str | dict[str, Any] | np.ndarray]] = []
         np_dtype = np.float32
 
         # Load xml data
@@ -66,7 +73,7 @@ class SVT(VisionDataset):
         )
         xml_root = xml_tree.getroot()
 
-        for image in tqdm(iterable=xml_root, desc="Unpacking SVT", total=len(xml_root)):
+        for image in tqdm(iterable=xml_root, desc="Preparing and Loading SVT", total=len(xml_root)):
             name, _, _, _resolution, rectangles = image
 
             # File existence check
@@ -108,6 +115,8 @@ class SVT(VisionDataset):
                 for crop, label in zip(crops, labels):
                     if crop.shape[0] > 0 and crop.shape[1] > 0 and len(label) > 0:
                         self.data.append((crop, label))
+            elif detection_task:
+                self.data.append((name.text, boxes))
             else:
                 self.data.append((name.text, dict(boxes=boxes, labels=labels)))
 

doctr/datasets/synthtext.py

@@ -1,11 +1,11 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
 
 import glob
 import os
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import numpy as np
 from PIL import Image
@@ -31,10 +31,10 @@ class SynthText(VisionDataset):
     >>> img, target = train_set[0]
 
     Args:
-    ----
         train: whether the subset should be the training one
         use_polygons: whether polygons should be considered as rotated bounding box (instead of straight ones)
         recognition_task: whether the dataset should be used for recognition task
+        detection_task: whether the dataset should be used for detection task
         **kwargs: keyword arguments from `VisionDataset`.
     """
 
@@ -46,6 +46,7 @@ class SynthText(VisionDataset):
         train: bool = True,
         use_polygons: bool = False,
         recognition_task: bool = False,
+        detection_task: bool = False,
         **kwargs: Any,
     ) -> None:
         super().__init__(
@@ -56,8 +57,14 @@ class SynthText(VisionDataset):
             pre_transforms=convert_target_to_relative if not recognition_task else None,
             **kwargs,
         )
+        if recognition_task and detection_task:
+            raise ValueError(
+                "`recognition_task` and `detection_task` cannot be set to True simultaneously. "
+                + "To get the whole dataset with boxes and labels leave both parameters to False."
+            )
+
         self.train = train
-        self.data: List[Tuple[Union[str, np.ndarray], Union[str, Dict[str, Any]]]] = []
+        self.data: list[tuple[str | np.ndarray, str | dict[str, Any] | np.ndarray]] = []
         np_dtype = np.float32
 
         # Load mat data
@@ -83,7 +90,7 @@ class SynthText(VisionDataset):
         del mat_data
 
         for img_path, word_boxes, txt in tqdm(
-            iterable=zip(paths, boxes, labels), desc="Unpacking SynthText", total=len(paths)
+            iterable=zip(paths, boxes, labels), desc="Preparing and Loading SynthText", total=len(paths)
         ):
             # File existence check
             if not os.path.exists(os.path.join(tmp_root, img_path[0])):
@@ -111,6 +118,8 @@ class SynthText(VisionDataset):
                             tmp_img = Image.fromarray(crop)
                             tmp_img.save(os.path.join(reco_folder_path, f"{reco_images_counter}.png"))
                             reco_images_counter += 1
+            elif detection_task:
+                self.data.append((img_path[0], np.asarray(word_boxes, dtype=np_dtype)))
             else:
                 self.data.append((img_path[0], dict(boxes=np.asarray(word_boxes, dtype=np_dtype), labels=labels)))
 

doctr/datasets/utils.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
@@ -6,10 +6,10 @@
 import string
 import unicodedata
 from collections.abc import Sequence
+from collections.abc import Sequence as SequenceType
 from functools import partial
 from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple, TypeVar, Union
-from typing import Sequence as SequenceType
+from typing import Any, TypeVar
 
 import numpy as np
 from PIL import Image
@@ -19,7 +19,15 @@ from doctr.utils.geometry import convert_to_relative_coords, extract_crops, extr
 
 from .vocabs import VOCABS
 
-__all__ = ["translate", "encode_string", "decode_sequence", "encode_sequences", "pre_transform_multiclass"]
+__all__ = [
+    "translate",
+    "encode_string",
+    "decode_sequence",
+    "encode_sequences",
+    "pre_transform_multiclass",
+    "crop_bboxes_from_image",
+    "convert_target_to_relative",
+]
 
 ImageTensor = TypeVar("ImageTensor")
 
@@ -32,13 +40,11 @@ def translate(
     """Translate a string input in a given vocabulary
 
     Args:
-    ----
         input_string: input string to translate
         vocab_name: vocabulary to use (french, latin, ...)
         unknown_char: unknown character for non-translatable characters
 
     Returns:
-    -------
         A string translated in a given vocab
     """
     if VOCABS.get(vocab_name) is None:
@@ -63,16 +69,14 @@ def translate(
 def encode_string(
     input_string: str,
     vocab: str,
-) -> List[int]:
+) -> list[int]:
     """Given a predefined mapping, encode the string to a sequence of numbers
 
     Args:
-    ----
         input_string: string to encode
         vocab: vocabulary (string), the encoding is given by the indexing of the character sequence
 
     Returns:
-    -------
         A list encoding the input_string
     """
     try:
@@ -85,18 +89,16 @@ def encode_string(
 
 
 def decode_sequence(
-    input_seq: Union[np.ndarray, SequenceType[int]],
+    input_seq: np.ndarray | SequenceType[int],
     mapping: str,
 ) -> str:
     """Given a predefined mapping, decode the sequence of numbers to a string
 
     Args:
-    ----
         input_seq: array to decode
         mapping: vocabulary (string), the encoding is given by the indexing of the character sequence
 
     Returns:
-    -------
         A string, decoded from input_seq
     """
     if not isinstance(input_seq, (Sequence, np.ndarray)):
@@ -108,18 +110,17 @@ def decode_sequence(
 
 
 def encode_sequences(
-    sequences: List[str],
+    sequences: list[str],
     vocab: str,
-    target_size: Optional[int] = None,
+    target_size: int | None = None,
     eos: int = -1,
-    sos: Optional[int] = None,
-    pad: Optional[int] = None,
+    sos: int | None = None,
+    pad: int | None = None,
     dynamic_seq_length: bool = False,
 ) -> np.ndarray:
     """Encode character sequences using a given vocab as mapping
 
     Args:
-    ----
         sequences: the list of character sequences of size N
         vocab: the ordered vocab to use for encoding
         target_size: maximum length of the encoded data
@@ -129,7 +130,6 @@ def encode_sequences(
         dynamic_seq_length: if `target_size` is specified, uses it as upper bound and enables dynamic sequence size
 
     Returns:
-    -------
         the padded encoded data as a tensor
     """
     if 0 <= eos < len(vocab):
@@ -169,21 +169,33 @@ def encode_sequences(
     return encoded_data
 
 
-def convert_target_to_relative(img: ImageTensor, target: Dict[str, Any]) -> Tuple[ImageTensor, Dict[str, Any]]:
-    target["boxes"] = convert_to_relative_coords(target["boxes"], get_img_shape(img))
+def convert_target_to_relative(
+    img: ImageTensor, target: np.ndarray | dict[str, Any]
+) -> tuple[ImageTensor, dict[str, Any] | np.ndarray]:
+    """Converts target to relative coordinates
+
+    Args:
+        img: tf.Tensor or torch.Tensor representing the image
+        target: target to convert to relative coordinates (boxes (N, 4) or polygons (N, 4, 2))
+
+    Returns:
+        The image and the target in relative coordinates
+    """
+    if isinstance(target, np.ndarray):
+        target = convert_to_relative_coords(target, get_img_shape(img))  # type: ignore[arg-type]
+    else:
+        target["boxes"] = convert_to_relative_coords(target["boxes"], get_img_shape(img))  # type: ignore[arg-type]
     return img, target
 
 
-def crop_bboxes_from_image(img_path: Union[str, Path], geoms: np.ndarray) -> List[np.ndarray]:
+def crop_bboxes_from_image(img_path: str | Path, geoms: np.ndarray) -> list[np.ndarray]:
     """Crop a set of bounding boxes from an image
 
     Args:
-    ----
         img_path: path to the image
         geoms: a array of polygons of shape (N, 4, 2) or of straight boxes of shape (N, 4)
 
     Returns:
-    -------
         a list of cropped images
     """
     with Image.open(img_path) as pil_img:
@@ -196,21 +208,19 @@ def crop_bboxes_from_image(img_path: Union[str, Path], geoms: np.ndarray) -> Lis
     raise ValueError("Invalid geometry format")
 
 
-def pre_transform_multiclass(img, target: Tuple[np.ndarray, List]) -> Tuple[np.ndarray, Dict[str, List]]:
+def pre_transform_multiclass(img, target: tuple[np.ndarray, list]) -> tuple[np.ndarray, dict[str, list]]:
     """Converts multiclass target to relative coordinates.
 
     Args:
-    ----
         img: Image
         target: tuple of target polygons and their classes names
 
     Returns:
-    -------
         Image and dictionary of boxes, with class names as keys
     """
     boxes = convert_to_relative_coords(target[0], get_img_shape(img))
     boxes_classes = target[1]
-    boxes_dict: Dict = {k: [] for k in sorted(set(boxes_classes))}
+    boxes_dict: dict = {k: [] for k in sorted(set(boxes_classes))}
     for k, poly in zip(boxes_classes, boxes):
         boxes_dict[k].append(poly)
     boxes_dict = {k: np.stack(v, axis=0) for k, v in boxes_dict.items()}

doctr/datasets/vocabs.py

@@ -1,15 +1,14 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
 
 import string
-from typing import Dict
 
 __all__ = ["VOCABS"]
 
 
-VOCABS: Dict[str, str] = {
+VOCABS: dict[str, str] = {
     "digits": string.digits,
     "ascii_letters": string.ascii_letters,
     "punctuation": string.punctuation,
@@ -20,11 +19,16 @@ VOCABS: Dict[str, str] = {
     "arabic_digits": "٠١٢٣٤٥٦٧٨٩",
     "arabic_diacritics": "ًٌٍَُِّْ",
     "arabic_punctuation": "؟؛«»—",
-    "hindi_letters": "अआइईउऊऋॠऌॡएऐओऔअंअःकखगघङचछजझञटठडढणतथदधनपफबभमयरलवशषसह",
+    "hindi_letters": "अआइईउऊऋॠऌॡएऐओऔंःकखगघङचछजझञटठडढणतथदधनपफबभमयरलवशषसह",
     "hindi_digits": "०१२३४५६७८९",
-    "hindi_punctuation": "।,?!:्ॐ॰॥॰",
+    "hindi_punctuation": "।,?!:्ॐ॰॥",
+    "gujarati_vowels": "અઆઇઈઉઊઋએઐઓ",
+    "gujarati_consonants": "ખગઘચછજઝઞટઠડઢણતથદધનપફબભમયરલવશસહળક્ષ",
+    "gujarati_digits": "૦૧૨૩૪૫૬૭૮૯",
+    "gujarati_punctuation": "૰ઽ◌ંઃ॥ૐ઼ઁ" + "૱",
     "bangla_letters": "অআইঈউঊঋএঐওঔকখগঘঙচছজঝঞটঠডঢণতথদধনপফবভমযরলশষসহ়ঽািীুূৃেৈোৌ্ৎংঃঁ",
     "bangla_digits": "০১২৩৪৫৬৭৮৯",
+    "generic_cyrillic_letters": "абвгдежзийклмнопрстуфхцчшщьюяАБВГДЕЖЗИЙКЛМНОПРСТУФХЦЧШЩЬЮЯ",
 }
 
 VOCABS["latin"] = VOCABS["digits"] + VOCABS["ascii_letters"] + VOCABS["punctuation"]
@@ -53,12 +57,22 @@ VOCABS["finnish"] = VOCABS["english"] + "äöÄÖ"
 VOCABS["swedish"] = VOCABS["english"] + "åäöÅÄÖ"
 VOCABS["vietnamese"] = (
     VOCABS["english"]
-    + "áàảạãăắằẳẵặâấầẩẫậéèẻẽẹêếềểễệóòỏõọôốồổộỗơớờởợỡúùủũụưứừửữựiíìỉĩịýỳỷỹỵ"
-    + "ÁÀẢẠÃĂẮẰẲẴẶÂẤẦẨẪẬÉÈẺẼẸÊẾỀỂỄỆÓÒỎÕỌÔỐỒỔỘỖƠỚỜỞỢỠÚÙỦŨỤƯỨỪỬỮỰIÍÌỈĨỊÝỲỶỸỴ"
+    + "áàảạãăắằẳẵặâấầẩẫậđéèẻẽẹêếềểễệóòỏõọôốồổộỗơớờởợỡúùủũụưứừửữựíìỉĩịýỳỷỹỵ"
+    + "ÁÀẢẠÃĂẮẰẲẴẶÂẤẦẨẪẬĐÉÈẺẼẸÊẾỀỂỄỆÓÒỎÕỌÔỐỒỔỘỖƠỚỜỞỢỠÚÙỦŨỤƯỨỪỬỮỰÍÌỈĨỊÝỲỶỸỴ"
 )
 VOCABS["hebrew"] = VOCABS["english"] + "אבגדהוזחטיכלמנסעפצקרשת" + "₪"
 VOCABS["hindi"] = VOCABS["hindi_letters"] + VOCABS["hindi_digits"] + VOCABS["hindi_punctuation"]
+VOCABS["gujarati"] = (
+    VOCABS["gujarati_vowels"]
+    + VOCABS["gujarati_consonants"]
+    + VOCABS["gujarati_digits"]
+    + VOCABS["gujarati_punctuation"]
+    + VOCABS["punctuation"]
+)
 VOCABS["bangla"] = VOCABS["bangla_letters"] + VOCABS["bangla_digits"]
+VOCABS["ukrainian"] = (
+    VOCABS["generic_cyrillic_letters"] + VOCABS["digits"] + VOCABS["punctuation"] + VOCABS["currency"] + "ґіїєҐІЇЄ₴"
+)
 VOCABS["multilingual"] = "".join(
     dict.fromkeys(
         VOCABS["french"]

doctr/datasets/wildreceipt.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
@@ -6,9 +6,10 @@
 import json
 import os
 from pathlib import Path
-from typing import Any, Dict, List, Tuple, Union
+from typing import Any
 
 import numpy as np
+from tqdm import tqdm
 
 from .datasets import AbstractDataset
 from .utils import convert_target_to_relative, crop_bboxes_from_image
@@ -17,9 +18,10 @@ __all__ = ["WILDRECEIPT"]
 
 
 class WILDRECEIPT(AbstractDataset):
-    """WildReceipt dataset from `"Spatial Dual-Modality Graph Reasoning for Key Information Extraction"
-        <https://arxiv.org/abs/2103.14470v1>`_ |
-    `repository <https://download.openmmlab.com/mmocr/data/wildreceipt.tar>`_.
+    """
+    WildReceipt dataset from `"Spatial Dual-Modality Graph Reasoning for Key Information Extraction"
+    <https://arxiv.org/abs/2103.14470v1>`_ |
+    `"repository" <https://download.openmmlab.com/mmocr/data/wildreceipt.tar>`_.
 
     .. image:: https://doctr-static.mindee.com/models?id=v0.7.0/wildreceipt-dataset.jpg&src=0
         :align: center
@@ -34,12 +36,12 @@ class WILDRECEIPT(AbstractDataset):
     >>> img, target = test_set[0]
 
     Args:
-    ----
         img_folder: folder with all the images of the dataset
         label_path: path to the annotations file of the dataset
         train: whether the subset should be the training one
         use_polygons: whether polygons should be considered as rotated bounding box (instead of straight ones)
         recognition_task: whether the dataset should be used for recognition task
+        detection_task: whether the dataset should be used for detection task
         **kwargs: keyword arguments from `AbstractDataset`.
     """
 
@@ -50,11 +52,19 @@ class WILDRECEIPT(AbstractDataset):
         train: bool = True,
         use_polygons: bool = False,
         recognition_task: bool = False,
+        detection_task: bool = False,
         **kwargs: Any,
     ) -> None:
         super().__init__(
             img_folder, pre_transforms=convert_target_to_relative if not recognition_task else None, **kwargs
         )
+        # Task check
+        if recognition_task and detection_task:
+            raise ValueError(
+                "`recognition_task` and `detection_task` cannot be set to True simultaneously. "
+                + "To get the whole dataset with boxes and labels leave both parameters to False."
+            )
+
         # File existence check
         if not os.path.exists(label_path) or not os.path.exists(img_folder):
             raise FileNotFoundError(f"unable to locate {label_path if not os.path.exists(label_path) else img_folder}")
@@ -62,15 +72,18 @@ class WILDRECEIPT(AbstractDataset):
         tmp_root = img_folder
         self.train = train
         np_dtype = np.float32
-        self.data: List[Tuple[Union[str, Path, np.ndarray], Union[str, Dict[str, Any]]]] = []
+        self.data: list[tuple[str | Path | np.ndarray, str | dict[str, Any] | np.ndarray]] = []
 
         with open(label_path, "r") as file:
             data = file.read()
         # Split the text file into separate JSON strings
         json_strings = data.strip().split("\n")
-        box: Union[List[float], np.ndarray]
-        _targets = []
-        for json_string in json_strings:
+        box: list[float] | np.ndarray
+
+        for json_string in tqdm(
+            iterable=json_strings, desc="Preparing and Loading WILDRECEIPT", total=len(json_strings)
+        ):
+            _targets = []
             json_data = json.loads(json_string)
             img_path = json_data["file_name"]
             annotations = json_data["annotations"]
@@ -100,6 +113,8 @@ class WILDRECEIPT(AbstractDataset):
                 for crop, label in zip(crops, list(text_targets)):
                     if label and " " not in label:
                         self.data.append((crop, label))
+            elif detection_task:
+                self.data.append((img_path, np.asarray(box_targets, dtype=int).clip(min=0)))
             else:
                 self.data.append((
                     img_path,

doctr/file_utils.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2024, Mindee.
+# Copyright (C) 2021-2025, 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.
@@ -9,7 +9,6 @@ import importlib.metadata
 import importlib.util
 import logging
 import os
-from typing import Optional
 
 CLASS_NAME: str = "words"
 
@@ -35,6 +34,20 @@ else:  # pragma: no cover
     logging.info("Disabling PyTorch because USE_TF is set")
     _torch_available = False
 
+# Compatibility fix to make sure tensorflow.keras stays at Keras 2
+if "TF_USE_LEGACY_KERAS" not in os.environ:
+    os.environ["TF_USE_LEGACY_KERAS"] = "1"
+
+elif os.environ["TF_USE_LEGACY_KERAS"] != "1":
+    raise ValueError(
+        "docTR is only compatible with Keras 2, but you have explicitly set `TF_USE_LEGACY_KERAS` to `0`. "
+    )
+
+
+def ensure_keras_v2() -> None:  # pragma: no cover
+    if not os.environ.get("TF_USE_LEGACY_KERAS") == "1":
+        os.environ["TF_USE_LEGACY_KERAS"] = "1"
+
 
 if USE_TF in ENV_VARS_TRUE_AND_AUTO_VALUES and USE_TORCH not in ENV_VARS_TRUE_VALUES:
     _tf_available = importlib.util.find_spec("tensorflow") is not None
@@ -65,6 +78,8 @@ if USE_TF in ENV_VARS_TRUE_AND_AUTO_VALUES and USE_TORCH not in ENV_VARS_TRUE_VA
             _tf_available = False
         else:
             logging.info(f"TensorFlow version {_tf_version} available.")
+            ensure_keras_v2()
+
 else:  # pragma: no cover
     logging.info("Disabling Tensorflow because USE_TORCH is set")
     _tf_available = False
@@ -77,12 +92,11 @@ if not _torch_available and not _tf_available:  # pragma: no cover
     )
 
 
-def requires_package(name: str, extra_message: Optional[str] = None) -> None:  # pragma: no cover
+def requires_package(name: str, extra_message: str | None = None) -> None:  # pragma: no cover
     """
     package requirement helper
 
     Args:
-    ----
         name: name of the package
         extra_message: additional message to display if the package is not found
     """