python-doctr 0.12.0__py3-none-any.whl → 1.0.1__py3-none-any.whl

doctr/models/classification/resnet/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *  # type: ignore[assignment]
+from .pytorch import *

doctr/models/classification/resnet/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/classification/textnet/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *
+from .pytorch import *

doctr/models/classification/textnet/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -11,7 +11,7 @@ from torch import nn
 
 from doctr.datasets import VOCABS
 
-from ...modules.layers.pytorch import FASTConvLayer
+from ...modules.layers import FASTConvLayer
 from ...utils import conv_sequence_pt, load_pretrained_params
 
 __all__ = ["textnet_tiny", "textnet_small", "textnet_base"]

doctr/models/classification/vgg/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *
+from .pytorch import *

doctr/models/classification/vgg/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/classification/vip/__init__.py

@@ -1,4 +1 @@
-from doctr.file_utils import is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
+from .pytorch import *

doctr/models/classification/vip/layers/__init__.py

@@ -1,4 +1 @@
-from doctr.file_utils import is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
+from .pytorch import *

doctr/models/classification/vip/layers/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -433,7 +433,7 @@ class LePEAttention(nn.Module):
         Returns:
             A float tensor of shape (b, h, w, c).
         """
-        b_merged = int(img_splits_hw.shape[0] / (h * w / h_sp / w_sp))
+        b_merged = img_splits_hw.shape[0] // ((h * w) // (h_sp * w_sp))
         img = img_splits_hw.view(b_merged, h // h_sp, w // w_sp, h_sp, w_sp, -1)
         # contiguous() required to ensure the tensor has a contiguous memory layout
         # after permute, allowing the subsequent view operation to work correctly.

doctr/models/classification/vip/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/classification/vit/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *
+from .pytorch import *

doctr/models/classification/vit/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -11,9 +11,9 @@ from torch import nn
 
 from doctr.datasets import VOCABS
 from doctr.models.modules.transformer import EncoderBlock
-from doctr.models.modules.vision_transformer.pytorch import PatchEmbedding
+from doctr.models.modules.vision_transformer import PatchEmbedding
 
-from ...utils.pytorch import load_pretrained_params
+from ...utils import load_pretrained_params
 
 __all__ = ["vit_s", "vit_b"]
 

doctr/models/classification/zoo.py

@@ -1,11 +1,11 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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
 
-from doctr.file_utils import is_tf_available, is_torch_available
+from doctr.models.utils import _CompiledModule
 
 from .. import classification
 from ..preprocessor import PreProcessor
@@ -30,11 +30,10 @@ ARCHS: list[str] = [
     "vgg16_bn_r",
     "vit_s",
     "vit_b",
+    "vip_tiny",
+    "vip_base",
 ]
 
-if is_torch_available():
-    ARCHS.extend(["vip_tiny", "vip_base"])
-
 ORIENTATION_ARCHS: list[str] = ["mobilenet_v3_small_crop_orientation", "mobilenet_v3_small_page_orientation"]
 
 
@@ -52,12 +51,8 @@ def _orientation_predictor(
         # Load directly classifier from backbone
         _model = classification.__dict__[arch](pretrained=pretrained)
     else:
-        allowed_archs = [classification.MobileNetV3]
-        if is_torch_available():
-            # Adding the type for torch compiled models to the allowed architectures
-            from doctr.models.utils import _CompiledModule
-
-            allowed_archs.append(_CompiledModule)
+        # Adding the type for torch compiled models to the allowed architectures
+        allowed_archs = [classification.MobileNetV3, _CompiledModule]
 
         if not isinstance(arch, tuple(allowed_archs)):
             raise ValueError(f"unknown architecture: {type(arch)}")
@@ -66,7 +61,7 @@ def _orientation_predictor(
     kwargs["mean"] = kwargs.get("mean", _model.cfg["mean"])
     kwargs["std"] = kwargs.get("std", _model.cfg["std"])
     kwargs["batch_size"] = kwargs.get("batch_size", 128 if model_type == "crop" else 4)
-    input_shape = _model.cfg["input_shape"][:-1] if is_tf_available() else _model.cfg["input_shape"][1:]
+    input_shape = _model.cfg["input_shape"][1:]
     predictor = OrientationPredictor(
         PreProcessor(input_shape, preserve_aspect_ratio=True, symmetric_pad=True, **kwargs), _model
     )

doctr/models/core.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/detection/_utils/__init__.py

@@ -1,7 +1,2 @@
-from doctr.file_utils import is_tf_available, is_torch_available
 from .base import *
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *
+from .pytorch import *

doctr/models/detection/_utils/base.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/detection/_utils/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.

doctr/models/detection/core.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -53,7 +53,7 @@ class DetectionPostProcessor(NestedObject):
 
         else:
             mask: np.ndarray = np.zeros((h, w), np.int32)
-            cv2.fillPoly(mask, [points.astype(np.int32)], 1.0)  # type: ignore[call-overload]
+            cv2.fillPoly(mask, [points.astype(np.int32)], 1.0)
             product = pred * mask
             return np.sum(product) / np.count_nonzero(product)
 

doctr/models/detection/differentiable_binarization/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *  # type: ignore[assignment]
+from .pytorch import *

doctr/models/detection/differentiable_binarization/base.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -224,7 +224,7 @@ class _DBNet:
         padded_polygon: np.ndarray = np.array(padding.Execute(distance)[0])
 
         # Fill the mask with 1 on the new padded polygon
-        cv2.fillPoly(mask, [padded_polygon.astype(np.int32)], 1.0)  # type: ignore[call-overload]
+        cv2.fillPoly(mask, [padded_polygon.astype(np.int32)], 1.0)
 
         # Get min/max to recover polygon after distance computation
         xmin = padded_polygon[:, 0].min()
@@ -269,7 +269,6 @@ class _DBNet:
         self,
         target: list[dict[str, np.ndarray]],
         output_shape: tuple[int, int, int],
-        channels_last: bool = True,
     ) -> tuple[np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
         if any(t.dtype != np.float32 for tgt in target for t in tgt.values()):
             raise AssertionError("the expected dtype of target 'boxes' entry is 'np.float32'.")
@@ -280,10 +279,8 @@ class _DBNet:
 
         h: int
         w: int
-        if channels_last:
-            h, w, num_classes = output_shape
-        else:
-            num_classes, h, w = output_shape
+
+        num_classes, h, w = output_shape
         target_shape = (len(target), num_classes, h, w)
 
         seg_target: np.ndarray = np.zeros(target_shape, dtype=np.uint8)
@@ -343,17 +340,12 @@ class _DBNet:
                     if shrunken.shape[0] <= 2 or not Polygon(shrunken).is_valid:
                         seg_mask[idx, class_idx, box[1] : box[3] + 1, box[0] : box[2] + 1] = False
                         continue
-                    cv2.fillPoly(seg_target[idx, class_idx], [shrunken.astype(np.int32)], 1.0)  # type: ignore[call-overload]
+                    cv2.fillPoly(seg_target[idx, class_idx], [shrunken.astype(np.int32)], 1.0)
 
                     # Draw on both thresh map and thresh mask
                     poly, thresh_target[idx, class_idx], thresh_mask[idx, class_idx] = self.draw_thresh_map(
                         poly, thresh_target[idx, class_idx], thresh_mask[idx, class_idx]
                     )
-        if channels_last:
-            seg_target = seg_target.transpose((0, 2, 3, 1))
-            seg_mask = seg_mask.transpose((0, 2, 3, 1))
-            thresh_target = thresh_target.transpose((0, 2, 3, 1))
-            thresh_mask = thresh_mask.transpose((0, 2, 3, 1))
 
         thresh_target = thresh_target.astype(input_dtype) * (self.thresh_max - self.thresh_min) + self.thresh_min
 

doctr/models/detection/differentiable_binarization/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -215,7 +215,7 @@ class DBNet(_DBNet, nn.Module):
 
         if target is None or return_preds:
             # Disable for torch.compile compatibility
-            @torch.compiler.disable  # type: ignore[attr-defined]
+            @torch.compiler.disable
             def _postprocess(prob_map: torch.Tensor) -> list[dict[str, Any]]:
                 return [
                     dict(zip(self.class_names, preds))
@@ -261,7 +261,7 @@ class DBNet(_DBNet, nn.Module):
         prob_map = torch.sigmoid(out_map)
         thresh_map = torch.sigmoid(thresh_map)
 
-        targets = self.build_target(target, out_map.shape[1:], False)  # type: ignore[arg-type]
+        targets = self.build_target(target, out_map.shape[1:])  # type: ignore[arg-type]
 
         seg_target, seg_mask = torch.from_numpy(targets[0]), torch.from_numpy(targets[1])
         seg_target, seg_mask = seg_target.to(out_map.device), seg_mask.to(out_map.device)
@@ -285,7 +285,7 @@ class DBNet(_DBNet, nn.Module):
                 dice_map = torch.softmax(out_map, dim=1)
             else:
                 # compute binary map instead
-                dice_map = 1 / (1 + torch.exp(-50.0 * (prob_map - thresh_map)))  # type: ignore[assignment]
+                dice_map = 1 / (1 + torch.exp(-50.0 * (prob_map - thresh_map)))
             # Class reduced
             inter = (seg_mask * dice_map * seg_target).sum((0, 2, 3))
             cardinality = (seg_mask * (dice_map + seg_target)).sum((0, 2, 3))

doctr/models/detection/fast/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *  # type: ignore[assignment]
+from .pytorch import *

doctr/models/detection/fast/base.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -153,14 +153,12 @@ class _FAST(BaseModel):
         self,
         target: list[dict[str, np.ndarray]],
         output_shape: tuple[int, int, int],
-        channels_last: bool = True,
     ) -> tuple[np.ndarray, np.ndarray, np.ndarray]:
         """Build the target, and it's mask to be used from loss computation.
 
         Args:
             target: target coming from dataset
             output_shape: shape of the output of the model without batch_size
-            channels_last: whether channels are last or not
 
         Returns:
             the new formatted target, mask and shrunken text kernel
@@ -172,10 +170,8 @@ class _FAST(BaseModel):
 
         h: int
         w: int
-        if channels_last:
-            h, w, num_classes = output_shape
-        else:
-            num_classes, h, w = output_shape
+
+        num_classes, h, w = output_shape
         target_shape = (len(target), num_classes, h, w)
 
         seg_target: np.ndarray = np.zeros(target_shape, dtype=np.uint8)
@@ -235,14 +231,8 @@ class _FAST(BaseModel):
                     if shrunken.shape[0] <= 2 or not Polygon(shrunken).is_valid:
                         seg_mask[idx, class_idx, box[1] : box[3] + 1, box[0] : box[2] + 1] = False
                         continue
-                    cv2.fillPoly(shrunken_kernel[idx, class_idx], [shrunken.astype(np.int32)], 1.0)  # type: ignore[call-overload]
+                    cv2.fillPoly(shrunken_kernel[idx, class_idx], [shrunken.astype(np.int32)], 1.0)
                     # draw the original polygon on the segmentation target
-                    cv2.fillPoly(seg_target[idx, class_idx], [poly.astype(np.int32)], 1.0)  # type: ignore[call-overload]
-
-        # Don't forget to switch back to channel last if Tensorflow is used
-        if channels_last:
-            seg_target = seg_target.transpose((0, 2, 3, 1))
-            seg_mask = seg_mask.transpose((0, 2, 3, 1))
-            shrunken_kernel = shrunken_kernel.transpose((0, 2, 3, 1))
+                    cv2.fillPoly(seg_target[idx, class_idx], [poly.astype(np.int32)], 1.0)
 
         return seg_target, seg_mask, shrunken_kernel

doctr/models/detection/fast/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -206,7 +206,7 @@ class FAST(_FAST, nn.Module):
 
         if target is None or return_preds:
             # Disable for torch.compile compatibility
-            @torch.compiler.disable  # type: ignore[attr-defined]
+            @torch.compiler.disable
             def _postprocess(prob_map: torch.Tensor) -> list[dict[str, Any]]:
                 return [
                     dict(zip(self.class_names, preds))
@@ -238,7 +238,7 @@ class FAST(_FAST, nn.Module):
         Returns:
             A loss tensor
         """
-        targets = self.build_target(target, out_map.shape[1:], False)  # type: ignore[arg-type]
+        targets = self.build_target(target, out_map.shape[1:])  # type: ignore[arg-type]
 
         seg_target, seg_mask = torch.from_numpy(targets[0]), torch.from_numpy(targets[1])
         shrunken_kernel = torch.from_numpy(targets[2]).to(out_map.device)
@@ -303,7 +303,7 @@ def reparameterize(model: FAST | nn.Module) -> FAST:
 
     for module in model.modules():
         if hasattr(module, "reparameterize_layer"):
-            module.reparameterize_layer()
+            module.reparameterize_layer()  # type: ignore[operator]
 
     for name, child in model.named_children():
         if isinstance(child, nn.BatchNorm2d):
@@ -315,7 +315,7 @@ def reparameterize(model: FAST | nn.Module) -> FAST:
 
             factor = child.weight / torch.sqrt(child.running_var + child.eps)  # type: ignore
             last_conv.weight = nn.Parameter(conv_w * factor.reshape([last_conv.out_channels, 1, 1, 1]))
-            last_conv.bias = nn.Parameter((conv_b - child.running_mean) * factor + child.bias)
+            last_conv.bias = nn.Parameter((conv_b - child.running_mean) * factor + child.bias)  # type: ignore[operator]
             model._modules[last_conv_name] = last_conv  # type: ignore[index]
             model._modules[name] = nn.Identity()
             last_conv = None

doctr/models/detection/linknet/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *  # type: ignore[assignment]
+from .pytorch import *

doctr/models/detection/linknet/base.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -156,14 +156,12 @@ class _LinkNet(BaseModel):
         self,
         target: list[dict[str, np.ndarray]],
         output_shape: tuple[int, int, int],
-        channels_last: bool = True,
     ) -> tuple[np.ndarray, np.ndarray]:
         """Build the target, and it's mask to be used from loss computation.
 
         Args:
             target: target coming from dataset
             output_shape: shape of the output of the model without batch_size
-            channels_last: whether channels are last or not
 
         Returns:
             the new formatted target and the mask
@@ -175,10 +173,8 @@ class _LinkNet(BaseModel):
 
         h: int
         w: int
-        if channels_last:
-            h, w, num_classes = output_shape
-        else:
-            num_classes, h, w = output_shape
+
+        num_classes, h, w = output_shape
         target_shape = (len(target), num_classes, h, w)
 
         seg_target: np.ndarray = np.zeros(target_shape, dtype=np.uint8)
@@ -237,11 +233,6 @@ class _LinkNet(BaseModel):
                     if shrunken.shape[0] <= 2 or not Polygon(shrunken).is_valid:
                         seg_mask[idx, class_idx, box[1] : box[3] + 1, box[0] : box[2] + 1] = False
                         continue
-                    cv2.fillPoly(seg_target[idx, class_idx], [shrunken.astype(np.int32)], 1.0)  # type: ignore[call-overload]
-
-        # Don't forget to switch back to channel last if Tensorflow is used
-        if channels_last:
-            seg_target = seg_target.transpose((0, 2, 3, 1))
-            seg_mask = seg_mask.transpose((0, 2, 3, 1))
+                    cv2.fillPoly(seg_target[idx, class_idx], [shrunken.astype(np.int32)], 1.0)
 
         return seg_target, seg_mask

doctr/models/detection/linknet/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -193,7 +193,7 @@ class LinkNet(nn.Module, _LinkNet):
 
         if target is None or return_preds:
             # Disable for torch.compile compatibility
-            @torch.compiler.disable  # type: ignore[attr-defined]
+            @torch.compiler.disable
             def _postprocess(prob_map: torch.Tensor) -> list[dict[str, Any]]:
                 return [
                     dict(zip(self.class_names, preds))
@@ -230,7 +230,7 @@ class LinkNet(nn.Module, _LinkNet):
         Returns:
             A loss tensor
         """
-        _target, _mask = self.build_target(target, out_map.shape[1:], False)  # type: ignore[arg-type]
+        _target, _mask = self.build_target(target, out_map.shape[1:])  # type: ignore[arg-type]
 
         seg_target, seg_mask = torch.from_numpy(_target).to(dtype=out_map.dtype), torch.from_numpy(_mask)
         seg_target, seg_mask = seg_target.to(out_map.device), seg_mask.to(out_map.device)

doctr/models/detection/predictor/__init__.py

@@ -1,6 +1 @@
-from doctr.file_utils import is_tf_available, is_torch_available
-
-if is_torch_available():
-    from .pytorch import *
-elif is_tf_available():
-    from .tensorflow import *  # type: ignore[assignment]
+from .pytorch import *

doctr/models/detection/predictor/pytorch.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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.
@@ -36,7 +36,7 @@ class DetectionPredictor(nn.Module):
     @torch.inference_mode()
     def forward(
         self,
-        pages: list[np.ndarray | torch.Tensor],
+        pages: list[np.ndarray],
         return_maps: bool = False,
         **kwargs: Any,
     ) -> list[dict[str, np.ndarray]] | tuple[list[dict[str, np.ndarray]], list[np.ndarray]]:

doctr/models/detection/zoo.py

@@ -1,11 +1,11 @@
-# Copyright (C) 2021-2025, Mindee.
+# Copyright (C) 2021-2026, 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
 
-from doctr.file_utils import is_tf_available, is_torch_available
+from doctr.models.utils import _CompiledModule
 
 from .. import detection
 from ..detection.fast import reparameterize
@@ -16,30 +16,17 @@ __all__ = ["detection_predictor"]
 
 ARCHS: list[str]
 
-
-if is_tf_available():
-    ARCHS = [
-        "db_resnet50",
-        "db_mobilenet_v3_large",
-        "linknet_resnet18",
-        "linknet_resnet34",
-        "linknet_resnet50",
-        "fast_tiny",
-        "fast_small",
-        "fast_base",
-    ]
-elif is_torch_available():
-    ARCHS = [
-        "db_resnet34",
-        "db_resnet50",
-        "db_mobilenet_v3_large",
-        "linknet_resnet18",
-        "linknet_resnet34",
-        "linknet_resnet50",
-        "fast_tiny",
-        "fast_small",
-        "fast_base",
-    ]
+ARCHS = [
+    "db_resnet34",
+    "db_resnet50",
+    "db_mobilenet_v3_large",
+    "linknet_resnet18",
+    "linknet_resnet34",
+    "linknet_resnet50",
+    "fast_tiny",
+    "fast_small",
+    "fast_base",
+]
 
 
 def _predictor(arch: Any, pretrained: bool, assume_straight_pages: bool = True, **kwargs: Any) -> DetectionPredictor:
@@ -56,12 +43,8 @@ def _predictor(arch: Any, pretrained: bool, assume_straight_pages: bool = True,
         if isinstance(_model, detection.FAST):
             _model = reparameterize(_model)
     else:
-        allowed_archs = [detection.DBNet, detection.LinkNet, detection.FAST]
-        if is_torch_available():
-            # Adding the type for torch compiled models to the allowed architectures
-            from doctr.models.utils import _CompiledModule
-
-            allowed_archs.append(_CompiledModule)
+        # Adding the type for torch compiled models to the allowed architectures
+        allowed_archs = [detection.DBNet, detection.LinkNet, detection.FAST, _CompiledModule]
 
         if not isinstance(arch, tuple(allowed_archs)):
             raise ValueError(f"unknown architecture: {type(arch)}")
@@ -76,7 +59,7 @@ def _predictor(arch: Any, pretrained: bool, assume_straight_pages: bool = True,
     kwargs["std"] = kwargs.get("std", _model.cfg["std"])
     kwargs["batch_size"] = kwargs.get("batch_size", 2)
     predictor = DetectionPredictor(
-        PreProcessor(_model.cfg["input_shape"][:-1] if is_tf_available() else _model.cfg["input_shape"][1:], **kwargs),
+        PreProcessor(_model.cfg["input_shape"][1:], **kwargs),
         _model,
     )
     return predictor