[models] Add benchmark fast_tiny and reparameterize by default

docs/source/using_doctr/using_models.rst

@@ -67,7 +67,7 @@ For a comprehensive comparison, we have compiled a detailed benchmark on publicl
 +----------------+---------------------------------+-----------------+---------------+------------+---------------+------------+---------------+--------------------+
 | PyTorch        | linknet_resnet50                | (1024, 1024, 3) | 28.8 M        | 81.78      | 82.47         | 87.29      | 85.54         | 1.0                |
 +----------------+---------------------------------+-----------------+---------------+------------+---------------+------------+---------------+--------------------+
-| PyTorch        | fast_tiny                       | (1024, 1024, 3) | 13.5 M (8.5M) |            |               |            |               | 0.7 (0.4)          |
+| PyTorch        | fast_tiny                       | (1024, 1024, 3) | 13.5 M (8.5M) | 84.90      | 85.04         | 93.73      | 76.26         | 0.7 (0.4)          |
 +----------------+---------------------------------+-----------------+---------------+------------+---------------+------------+---------------+--------------------+
 | PyTorch        | fast_small                      | (1024, 1024, 3) | 14.7 M (9.7M) |            |               |            |               | 0.7 (0.5)          |
 +----------------+---------------------------------+-----------------+---------------+------------+---------------+------------+---------------+--------------------+

doctr/models/detection/fast/base.py

@@ -31,7 +31,7 @@ class FASTPostProcessor(DetectionPostProcessor):
 
     def __init__(
         self,
-        bin_thresh: float = 0.3,
+        bin_thresh: float = 0.1,
         box_thresh: float = 0.1,
         assume_straight_pages: bool = True,
     ) -> None:

doctr/models/detection/fast/pytorch.py

@@ -119,7 +119,7 @@ class FAST(_FAST, nn.Module):
     def __init__(
         self,
         feat_extractor: IntermediateLayerGetter,
-        bin_thresh: float = 0.3,
+        bin_thresh: float = 0.1,
         box_thresh: float = 0.1,
         dropout_prob: float = 0.1,
         pooling_size: int = 4,  # different from paper performs better on close text-rich images

doctr/models/detection/fast/tensorflow.py

@@ -122,7 +122,7 @@ class FAST(_FAST, keras.Model, NestedObject):
     def __init__(
         self,
         feature_extractor: IntermediateLayerGetter,
-        bin_thresh: float = 0.3,
+        bin_thresh: float = 0.1,
         box_thresh: float = 0.1,
         dropout_prob: float = 0.1,
         pooling_size: int = 4,  # different from paper performs better on close text-rich images

doctr/models/detection/zoo.py

@@ -8,6 +8,7 @@
 from doctr.file_utils import is_tf_available, is_torch_available
 
 from .. import detection
+from ..detection.fast import reparameterize
 from ..preprocessor import PreProcessor
 from .predictor import DetectionPredictor
 
@@ -51,6 +52,9 @@ def _predictor(arch: Any, pretrained: bool, assume_straight_pages: bool = True,
             pretrained_backbone=kwargs.get("pretrained_backbone", True),
             assume_straight_pages=assume_straight_pages,
         )
+        # Reparameterize FAST models by default to lower inference latency and memory usage
+        if isinstance(_model, detection.FAST):
+            _model = reparameterize(_model)
     else:
         if not isinstance(arch, (detection.DBNet, detection.LinkNet, detection.FAST)):
             raise ValueError(f"unknown architecture: {type(arch)}")