tf-models-nightly 2.12.0.dev20230522__py2.py3-none-any.whl → 2.12.0.dev20230524__py2.py3-none-any.whl

official/projects/qat/vision/modeling/heads/dense_prediction_heads.py

@@ -13,10 +13,7 @@
 # limitations under the License.
 
 """Contains definitions of dense prediction heads."""
-from __future__ import annotations
-
-import copy
-from typing import Any, Dict, List, Mapping, Optional, Union, Type
+from typing import List, Mapping, Union, Optional, Any, Dict
 
 # Import libraries
 
@@ -29,80 +26,6 @@ from official.projects.qat.vision.quantization import configs
 from official.projects.qat.vision.quantization import helper
 
 
-class SeparableConv2DQuantized(tf.keras.layers.Layer):
-  """Quantized SeperableConv2D."""
-
-  def __init__(self,
-               name: Optional[str] = None,
-               last_quantize: bool = False,
-               **conv_kwargs):
-    """Initializes a SeparableConv2DQuantized.
-
-    Args:
-      name: The name of the layer.
-      last_quantize: A `bool` indicates whether add quantization for the output.
-      **conv_kwargs: A keyword arguments to be used for conv and dwconv.
-    """
-
-    super().__init__(name=name)
-    self._conv_kwargs = copy.deepcopy(conv_kwargs)
-    self._name = name
-    self._last_quantize = last_quantize
-
-  def build(self, input_shape: Union[tf.TensorShape, List[tf.TensorShape]]):
-    """Creates the child layers of the layer."""
-    depthwise_conv2d_quantized = helper.quantize_wrapped_layer(
-        tf.keras.layers.DepthwiseConv2D,
-        configs.Default8BitConvQuantizeConfig(
-            ['depthwise_kernel'], [], True))
-    conv2d_quantized = helper.quantize_wrapped_layer(
-        tf.keras.layers.Conv2D,
-        configs.Default8BitConvQuantizeConfig(
-            ['kernel'], [], self._last_quantize))
-
-    dwconv_kwargs = self._conv_kwargs.copy()
-    # Depthwise conv input filters is always equal to output filters.
-    # This filters argument only needed for the point-wise conv2d op.
-    del dwconv_kwargs['filters']
-    dwconv_kwargs.update({
-        'activation': None,
-        'use_bias': False,
-    })
-    self.dw_conv = depthwise_conv2d_quantized(name='dw', **dwconv_kwargs)
-
-    conv_kwargs = self._conv_kwargs.copy()
-    conv_kwargs.update({
-        'kernel_size': (1, 1),
-        'strides': (1, 1),
-        'padding': 'valid',
-        'groups': 1,
-    })
-
-    self.conv = conv2d_quantized(name='pw', **conv_kwargs)
-
-  def call(self, inputs: tf.Tensor) -> tf.Tensor:
-    """Call the separable conv layer."""
-    x = self.dw_conv(inputs)
-    outputs = self.conv(x)
-    return outputs
-
-  def get_config(self) -> Dict[str, Any]:
-    """Returns the config of the layer."""
-    config = self._conv_kwargs.copy()
-    config.update({
-        'name': self._name,
-        'last_quantize': self._last_quantize,
-    })
-    return config
-
-  @classmethod
-  def from_config(
-      cls: Type[SeparableConv2DQuantized],
-      config: Dict[str, Any]) -> SeparableConv2DQuantized:
-    """Creates a layer from its config."""
-    return cls(**config)
-
-
 @tf.keras.utils.register_keras_serializable(package='Vision')
 class RetinaNetHeadQuantized(tf.keras.layers.Layer):
   """Creates a RetinaNet quantized head."""
@@ -201,7 +124,7 @@ class RetinaNetHeadQuantized(tf.keras.layers.Layer):
   def build(self, input_shape: Union[tf.TensorShape, List[tf.TensorShape]]):
     """Creates the variables of the head."""
     if self._config_dict['use_separable_conv']:
-      conv_op = SeparableConv2DQuantized
+      conv_op = helper.SeparableConv2DQuantized
     else:
       conv_op = helper.quantize_wrapped_layer(
           tf.keras.layers.Conv2D,

official/projects/qat/vision/quantization/helper.py

@@ -13,7 +13,11 @@
 # limitations under the License.
 
 """Quantization helpers."""
-from typing import Any, Dict
+
+from __future__ import annotations
+
+import copy
+from typing import Any, Dict, List, Optional, Type, Union
 
 import tensorflow as tf
 
@@ -22,20 +26,42 @@ from official.projects.qat.vision.quantization import configs
 
 
 _QUANTIZATION_WEIGHT_NAMES = [
-    'output_max', 'output_min', 'optimizer_step', 'kernel_min', 'kernel_max',
-    'add_three_min', 'add_three_max', 'divide_six_min', 'divide_six_max',
-    'depthwise_kernel_min', 'depthwise_kernel_max',
-    'reduce_mean_quantizer_vars_min', 'reduce_mean_quantizer_vars_max',
-    'quantize_layer_min', 'quantize_layer_max',
-    'quantize_layer_1_min', 'quantize_layer_1_max',
-    'quantize_layer_2_min', 'quantize_layer_2_max',
-    'quantize_layer_3_min', 'quantize_layer_3_max',
-    'post_activation_min', 'post_activation_max',
+    'output_max',
+    'output_min',
+    'optimizer_step',
+    'kernel_min',
+    'kernel_max',
+    'add_three_min',
+    'add_three_max',
+    'divide_six_min',
+    'divide_six_max',
+    'depthwise_kernel_min',
+    'depthwise_kernel_max',
+    'pointwise_kernel_min',
+    'pointwise_kernel_max',
+    'reduce_mean_quantizer_vars_min',
+    'reduce_mean_quantizer_vars_max',
+    'quantize_layer_min',
+    'quantize_layer_max',
+    'quantize_layer_1_min',
+    'quantize_layer_1_max',
+    'quantize_layer_2_min',
+    'quantize_layer_2_max',
+    'quantize_layer_3_min',
+    'quantize_layer_3_max',
+    'post_activation_min',
+    'post_activation_max',
 ]
 
 _ORIGINAL_WEIGHT_NAME = [
-    'kernel', 'depthwise_kernel', 'gamma', 'beta', 'moving_mean',
-    'moving_variance', 'bias'
+    'kernel',
+    'depthwise_kernel',
+    'pointwise_kernel',
+    'gamma',
+    'beta',
+    'moving_mean',
+    'moving_variance',
+    'bias',
 ]
 
 
@@ -141,6 +167,84 @@ def norm_by_activation(activation, norm_quantized, norm_no_quantized):
     return norm_no_quantized
 
 
+class SeparableConv2DQuantized(tf.keras.layers.Layer):
+  """Quantized SeperableConv2D."""
+
+  def __init__(
+      self,
+      name: Optional[str] = None,
+      last_quantize: bool = False,
+      **conv_kwargs,
+  ):
+    """Initializes a SeparableConv2DQuantized.
+
+    Args:
+      name: The name of the layer.
+      last_quantize: A `bool` indicates whether add quantization for the output.
+      **conv_kwargs: A keyword arguments to be used for conv and dwconv.
+    """
+
+    super().__init__(name=name)
+    self._conv_kwargs = copy.deepcopy(conv_kwargs)
+    self._name = name
+    self._last_quantize = last_quantize
+
+  def build(self, input_shape: Union[tf.TensorShape, List[tf.TensorShape]]):
+    """Creates the child layers of the layer."""
+    depthwise_conv2d_quantized = quantize_wrapped_layer(
+        tf.keras.layers.DepthwiseConv2D,
+        configs.Default8BitConvQuantizeConfig(['depthwise_kernel'], [], True),
+    )
+    conv2d_quantized = quantize_wrapped_layer(
+        tf.keras.layers.Conv2D,
+        configs.Default8BitConvQuantizeConfig(
+            ['kernel'], [], self._last_quantize
+        ),
+    )
+
+    dwconv_kwargs = self._conv_kwargs.copy()
+    # Depthwise conv input filters is always equal to output filters.
+    # This filters argument only needed for the point-wise conv2d op.
+    del dwconv_kwargs['filters']
+    dwconv_kwargs.update({
+        'activation': None,
+        'use_bias': False,
+    })
+    self.dw_conv = depthwise_conv2d_quantized(name='dw', **dwconv_kwargs)
+
+    conv_kwargs = self._conv_kwargs.copy()
+    conv_kwargs.update({
+        'kernel_size': (1, 1),
+        'strides': (1, 1),
+        'padding': 'valid',
+        'groups': 1,
+    })
+
+    self.conv = conv2d_quantized(name='pw', **conv_kwargs)
+
+  def call(self, inputs: tf.Tensor) -> tf.Tensor:
+    """Call the separable conv layer."""
+    x = self.dw_conv(inputs)
+    outputs = self.conv(x)
+    return outputs
+
+  def get_config(self) -> Dict[str, Any]:
+    """Returns the config of the layer."""
+    config = self._conv_kwargs.copy()
+    config.update({
+        'name': self._name,
+        'last_quantize': self._last_quantize,
+    })
+    return config
+
+  @classmethod
+  def from_config(
+      cls: Type[SeparableConv2DQuantized], config: Dict[str, Any]
+  ) -> SeparableConv2DQuantized:
+    """Creates a layer from its config."""
+    return cls(**config)
+
+
 Conv2DQuantized = quantize_wrapped_layer(
     tf.keras.layers.Conv2D,
     configs.Default8BitConvQuantizeConfig(['kernel'], ['activation'], False))

official/projects/yolo/modeling/backbones/darknet_test.py

@@ -82,7 +82,10 @@ class DarknetTest(parameterized.TestCase, tf.test.TestCase):
 
     with strategy.scope():
       network = darknet.Darknet(
-          model_id='darknet53', min_size=3, max_size=5, use_sync_bn=use_sync_bn
+          model_id='darknet53',
+          min_level=3,
+          max_level=5,
+          use_sync_bn=use_sync_bn,
       )
       _ = network(inputs)
 

{tf_models_nightly-2.12.0.dev20230522.dist-info → tf_models_nightly-2.12.0.dev20230524.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tf-models-nightly
-Version: 2.12.0.dev20230522
+Version: 2.12.0.dev20230524
 Summary: TensorFlow Official Models
 Home-page: https://github.com/tensorflow/models
 Author: Google Inc.

{tf_models_nightly-2.12.0.dev20230522.dist-info → tf_models_nightly-2.12.0.dev20230524.dist-info}/RECORD

@@ -637,7 +637,7 @@ official/projects/qat/vision/modeling/factory.py,sha256=R3pujxDYJXKN7FGOP2ow5X94
 official/projects/qat/vision/modeling/factory_test.py,sha256=inWex80tdR56kNQtsEpPEIUTEYH7xElcH6Rsr2xyGrI,9737
 official/projects/qat/vision/modeling/segmentation_model.py,sha256=00zq2AWwjzTJ_9wp8TGc8PY-0tHFwYTK4s9SPUsgHRY,3077
 official/projects/qat/vision/modeling/heads/__init__.py,sha256=tjLYHwSnUDvWNhlwGa5XXF8jzKY0_5R5DxVgEDbzz4U,743
-official/projects/qat/vision/modeling/heads/dense_prediction_heads.py,sha256=zKZ7DGyKSyxaTxzZ1ofuPqrBobwAGh4HX_5Us1d0_dU,18206
+official/projects/qat/vision/modeling/heads/dense_prediction_heads.py,sha256=cYAGkcG3q6IqrC8gFIaNIQ8LrxhOGs8dHOsxd5XkE-U,15779
 official/projects/qat/vision/modeling/heads/dense_prediction_heads_test.py,sha256=W2EUV9h46Rtx72w6lWzmtsUM_thgsUkAOWO0SX4L3F4,3980
 official/projects/qat/vision/modeling/layers/__init__.py,sha256=Xgv0erxZbpq2F_NSo4xwAhM4v6XEzzO522CAsP9VIRw,991
 official/projects/qat/vision/modeling/layers/nn_blocks.py,sha256=Gznf3bUDunLUmGD_9dXl-ylvwbVyOA01sBaFT1oBhYc,30118
@@ -654,7 +654,7 @@ official/projects/qat/vision/n_bit/schemes.py,sha256=-879p8-PBmBdBuSegkfqGnNmVhn
 official/projects/qat/vision/quantization/__init__.py,sha256=mXv8p6ih9kX5K-JXs9M30OtFK7oYcZ-pfC5OKg6nitY,643
 official/projects/qat/vision/quantization/configs.py,sha256=TJlF55rQ8j3WOIXK1qds0nZZPuG7fSwJA9319IQi4Ug,11684
 official/projects/qat/vision/quantization/configs_test.py,sha256=BybuVWfy1hHPWl6HYKLz6U_m_h40Z82A0HBAOq1sKAk,6797
-official/projects/qat/vision/quantization/helper.py,sha256=xQFwIV3bLRTXZFT-h8kSwJhsRvhi0eY50sHWQzT8sII,7199
+official/projects/qat/vision/quantization/helper.py,sha256=nvq6EVDEgwBoeYe6p5mXy_lrccpzOsS0uKXYHgposaA,9789
 official/projects/qat/vision/quantization/helper_test.py,sha256=0308fCkPFyGdt1FK2bEvngptzWCdb2g5pQCf0sg15LM,1928
 official/projects/qat/vision/quantization/layer_transforms.py,sha256=vr9C73HDPMEh6gmpPaBopGFKu_cDW3QUEg6mzgf4VYY,5057
 official/projects/qat/vision/quantization/schemes.py,sha256=v8NyDpCUKelzKvBjjyHjlUbzLZfHkA2LJPufXb3UXNM,3089
@@ -743,7 +743,7 @@ official/projects/yolo/modeling/yolo_model.py,sha256=qZsrAaH3F9o2d94Jdbldou-XF_0
 official/projects/yolo/modeling/yolov7_model.py,sha256=4uxBffJZ7J29dsUUCsP2_xhLZTVfsJV48j5Hc63H1Po,3205
 official/projects/yolo/modeling/backbones/__init__.py,sha256=1ToRMjre4mErL4Ek4_dMVxMjXNPossNXggV8fqbISao,609
 official/projects/yolo/modeling/backbones/darknet.py,sha256=uj7iQkEsVM1lgYYTdsh0s8HXRIYUcCARzzPqC5-s-fo,22015
-official/projects/yolo/modeling/backbones/darknet_test.py,sha256=AoaAofPCyT2uDHvACTwsNuaZv_0t2x6SI3k-08zZNlg,4725
+official/projects/yolo/modeling/backbones/darknet_test.py,sha256=dCynaTj8OE0D191EKnA4qEk3Qvu5MroPtcK-iZgwymk,4758
 official/projects/yolo/modeling/backbones/yolov7.py,sha256=PCuJAeLachDFfewKBb0kmh9Fy9QwzJAykYUK0zSmnhs,12875
 official/projects/yolo/modeling/backbones/yolov7_test.py,sha256=kutqiGaxf_prQglxSaihhdYDer8ZzQRHMl7w7z76znc,3049
 official/projects/yolo/modeling/decoders/__init__.py,sha256=1ToRMjre4mErL4Ek4_dMVxMjXNPossNXggV8fqbISao,609
@@ -1102,9 +1102,9 @@ tensorflow_models/__init__.py,sha256=Ciz_YBke6teb6y42QyQTUBDdXJAiV7Qdu1zOoZvYiKw
 tensorflow_models/tensorflow_models_test.py,sha256=Kz2y4V-rtBhZFFfKD2soCq52hviSfJVV1L2ztqS-9oM,1385
 tensorflow_models/nlp/__init__.py,sha256=3dULDpUBpDi9vljpXadq6oJrWH4y6z42Bz2d3hopYZw,807
 tensorflow_models/vision/__init__.py,sha256=4y77XkHaH8qLls3-6ta4tMp3Xj8CLbB0ihH91HsQ9z4,833
-tf_models_nightly-2.12.0.dev20230522.dist-info/AUTHORS,sha256=1dG3fXVu9jlo7bul8xuix5F5vOnczMk7_yWn4y70uw0,337
-tf_models_nightly-2.12.0.dev20230522.dist-info/LICENSE,sha256=WxeBS_DejPZQabxtfMOM_xn8qoZNJDQjrT7z2wG1I4U,11512
-tf_models_nightly-2.12.0.dev20230522.dist-info/METADATA,sha256=E1Nu7QMA-rikmDz9e1a7FeacwUGNGfGQ4B27Z_aqhaU,1393
-tf_models_nightly-2.12.0.dev20230522.dist-info/WHEEL,sha256=kGT74LWyRUZrL4VgLh6_g12IeVl_9u9ZVhadrgXZUEY,110
-tf_models_nightly-2.12.0.dev20230522.dist-info/top_level.txt,sha256=gum2FfO5R4cvjl2-QtP-S1aNmsvIZaFFT6VFzU0f4-g,33
-tf_models_nightly-2.12.0.dev20230522.dist-info/RECORD,,
+tf_models_nightly-2.12.0.dev20230524.dist-info/AUTHORS,sha256=1dG3fXVu9jlo7bul8xuix5F5vOnczMk7_yWn4y70uw0,337
+tf_models_nightly-2.12.0.dev20230524.dist-info/LICENSE,sha256=WxeBS_DejPZQabxtfMOM_xn8qoZNJDQjrT7z2wG1I4U,11512
+tf_models_nightly-2.12.0.dev20230524.dist-info/METADATA,sha256=-S1EbFFT1XhMFrxG6OG7NJJ3KVsCEqH1A0uu-crodxc,1393
+tf_models_nightly-2.12.0.dev20230524.dist-info/WHEEL,sha256=kGT74LWyRUZrL4VgLh6_g12IeVl_9u9ZVhadrgXZUEY,110
+tf_models_nightly-2.12.0.dev20230524.dist-info/top_level.txt,sha256=gum2FfO5R4cvjl2-QtP-S1aNmsvIZaFFT6VFzU0f4-g,33
+tf_models_nightly-2.12.0.dev20230524.dist-info/RECORD,,