keras-hub-nightly 0.16.1.dev202410100339__py3-none-any.whl → 0.16.1.dev202410110340__py3-none-any.whl

keras_hub/api/layers/__init__.py

@@ -52,6 +52,7 @@ from keras_hub.src.models.resnet.resnet_image_converter import (
 from keras_hub.src.models.sam.sam_image_converter import SAMImageConverter
 from keras_hub.src.models.sam.sam_mask_decoder import SAMMaskDecoder
 from keras_hub.src.models.sam.sam_prompt_encoder import SAMPromptEncoder
+from keras_hub.src.models.vgg.vgg_image_classifier import VGGImageConverter
 from keras_hub.src.models.whisper.whisper_audio_converter import (
     WhisperAudioConverter,
 )

keras_hub/api/models/__init__.py

@@ -299,6 +299,9 @@ from keras_hub.src.models.text_classifier_preprocessor import (
 from keras_hub.src.models.text_to_image import TextToImage
 from keras_hub.src.models.vgg.vgg_backbone import VGGBackbone
 from keras_hub.src.models.vgg.vgg_image_classifier import VGGImageClassifier
+from keras_hub.src.models.vgg.vgg_image_classifier import (
+    VGGImageClassifierPreprocessor,
+)
 from keras_hub.src.models.vit_det.vit_det_backbone import ViTDetBackbone
 from keras_hub.src.models.whisper.whisper_backbone import WhisperBackbone
 from keras_hub.src.models.whisper.whisper_tokenizer import WhisperTokenizer

keras_hub/src/models/vgg/__init__.py

@@ -0,0 +1 @@
+from keras_hub.src.models.vgg.vgg_backbone import VGGBackbone

keras_hub/src/models/vgg/vgg_backbone.py

@@ -47,12 +47,11 @@ class VGGBackbone(Backbone):
         image_shape=(None, None, 3),
         **kwargs,
     ):
-
         # === Functional Model ===
         img_input = keras.layers.Input(shape=image_shape)
         x = img_input
 
-        for stack_index in range(len(stackwise_num_repeats) - 1):
+        for stack_index in range(len(stackwise_num_repeats)):
             x = apply_vgg_block(
                 x=x,
                 num_layers=stackwise_num_repeats[stack_index],

keras_hub/src/models/vgg/vgg_image_classifier.py

@@ -1,11 +1,26 @@
 import keras
 
 from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.layers.preprocessing.image_converter import ImageConverter
 from keras_hub.src.models.image_classifier import ImageClassifier
+from keras_hub.src.models.image_classifier_preprocessor import (
+    ImageClassifierPreprocessor,
+)
 from keras_hub.src.models.task import Task
 from keras_hub.src.models.vgg.vgg_backbone import VGGBackbone
 
 
+@keras_hub_export("keras_hub.layers.VGGImageConverter")
+class VGGImageConverter(ImageConverter):
+    backbone_cls = VGGBackbone
+
+
+@keras_hub_export("keras_hub.models.VGGImageClassifierPreprocessor")
+class VGGImageClassifierPreprocessor(ImageClassifierPreprocessor):
+    backbone_cls = VGGBackbone
+    image_converter_cls = VGGImageConverter
+
+
 @keras_hub_export("keras_hub.models.VGGImageClassifier")
 class VGGImageClassifier(ImageClassifier):
     """VGG image classification task.
@@ -96,13 +111,14 @@ class VGGImageClassifier(ImageClassifier):
     """
 
     backbone_cls = VGGBackbone
+    preprocessor_cls = VGGImageClassifierPreprocessor
 
     def __init__(
         self,
         backbone,
         num_classes,
         preprocessor=None,
-        pooling="flatten",
+        pooling="avg",
         pooling_hidden_dim=4096,
         activation=None,
         dropout=0.0,
@@ -141,24 +157,46 @@ class VGGImageClassifier(ImageClassifier):
                 "Unknown `pooling` type. Polling should be either `'avg'` or "
                 f"`'max'`. Received: pooling={pooling}."
             )
-        self.output_dropout = keras.layers.Dropout(
-            dropout,
-            dtype=head_dtype,
-            name="output_dropout",
-        )
-        self.output_dense = keras.layers.Dense(
-            num_classes,
-            activation=activation,
-            dtype=head_dtype,
-            name="predictions",
+
+        self.head = keras.Sequential(
+            [
+                keras.layers.Conv2D(
+                    filters=4096,
+                    kernel_size=7,
+                    name="fc1",
+                    activation=activation,
+                    use_bias=True,
+                    padding="same",
+                ),
+                keras.layers.Dropout(
+                    rate=dropout,
+                    dtype=head_dtype,
+                    name="output_dropout",
+                ),
+                keras.layers.Conv2D(
+                    filters=4096,
+                    kernel_size=1,
+                    name="fc2",
+                    activation=activation,
+                    use_bias=True,
+                    padding="same",
+                ),
+                self.pooler,
+                keras.layers.Dense(
+                    num_classes,
+                    activation=activation,
+                    dtype=head_dtype,
+                    name="predictions",
+                ),
+            ],
+            name="head",
         )
 
         # === Functional Model ===
         inputs = self.backbone.input
         x = self.backbone(inputs)
-        x = self.pooler(x)
-        x = self.output_dropout(x)
-        outputs = self.output_dense(x)
+        outputs = self.head(x)
+
         # Skip the parent class functional model.
         Task.__init__(
             self,

keras_hub/src/utils/timm/convert_vgg.py

@@ -0,0 +1,85 @@
+from typing import Any
+
+import numpy as np
+
+from keras_hub.src.models.vgg.vgg_backbone import VGGBackbone
+from keras_hub.src.models.vgg.vgg_image_classifier import VGGImageClassifier
+
+backbone_cls = VGGBackbone
+
+
+REPEATS_BY_SIZE = {
+    "vgg11": [1, 1, 2, 2, 2],
+    "vgg13": [2, 2, 2, 2, 2],
+    "vgg16": [2, 2, 3, 3, 3],
+    "vgg19": [2, 2, 4, 4, 4],
+}
+
+
+def convert_backbone_config(timm_config):
+    architecture = timm_config["architecture"]
+    stackwise_num_repeats = REPEATS_BY_SIZE[architecture]
+    return dict(
+        stackwise_num_repeats=stackwise_num_repeats,
+        stackwise_num_filters=[64, 128, 256, 512, 512],
+    )
+
+
+def convert_conv2d(
+    model,
+    loader,
+    keras_layer_name: str,
+    hf_layer_name: str,
+):
+    loader.port_weight(
+        model.get_layer(keras_layer_name).kernel,
+        hf_weight_key=f"{hf_layer_name}.weight",
+        hook_fn=lambda x, _: np.transpose(x, (2, 3, 1, 0)),
+    )
+    loader.port_weight(
+        model.get_layer(keras_layer_name).bias,
+        hf_weight_key=f"{hf_layer_name}.bias",
+    )
+
+
+def convert_weights(
+    backbone: VGGBackbone,
+    loader,
+    timm_config: dict[Any],
+):
+    architecture = timm_config["architecture"]
+    stackwise_num_repeats = REPEATS_BY_SIZE[architecture]
+
+    hf_index_to_keras_layer_name = {}
+    layer_index = 0
+    for block_index, repeats_in_block in enumerate(stackwise_num_repeats):
+        for repeat_index in range(repeats_in_block):
+            hf_index = layer_index
+            layer_index += 2  # Conv + activation layers.
+            layer_name = f"block{block_index + 1}_conv{repeat_index + 1}"
+            hf_index_to_keras_layer_name[hf_index] = layer_name
+        layer_index += 1  # Pooling layer after blocks.
+
+    for hf_index, keras_layer_name in hf_index_to_keras_layer_name.items():
+        convert_conv2d(
+            backbone, loader, keras_layer_name, f"features.{hf_index}"
+        )
+
+
+def convert_head(
+    task: VGGImageClassifier,
+    loader,
+    timm_config: dict[Any],
+):
+    convert_conv2d(task.head, loader, "fc1", "pre_logits.fc1")
+    convert_conv2d(task.head, loader, "fc2", "pre_logits.fc2")
+
+    loader.port_weight(
+        task.head.get_layer("predictions").kernel,
+        hf_weight_key="head.fc.weight",
+        hook_fn=lambda x, _: np.transpose(np.squeeze(x)),
+    )
+    loader.port_weight(
+        task.head.get_layer("predictions").bias,
+        hf_weight_key="head.fc.bias",
+    )

keras_hub/src/utils/timm/preset_loader.py

@@ -5,6 +5,7 @@ from keras_hub.src.utils.preset_utils import PresetLoader
 from keras_hub.src.utils.preset_utils import jax_memory_cleanup
 from keras_hub.src.utils.timm import convert_densenet
 from keras_hub.src.utils.timm import convert_resnet
+from keras_hub.src.utils.timm import convert_vgg
 from keras_hub.src.utils.transformers.safetensor_utils import SafetensorLoader
 
 
@@ -16,6 +17,8 @@ class TimmPresetLoader(PresetLoader):
             self.converter = convert_resnet
         elif "densenet" in architecture:
             self.converter = convert_densenet
+        elif "vgg" in architecture:
+            self.converter = convert_vgg
         else:
             raise ValueError(
                 "KerasHub has no converter for timm models "

keras_hub/src/version_utils.py

@@ -1,7 +1,7 @@
 from keras_hub.src.api_export import keras_hub_export
 
 # Unique source of truth for the version number.
-__version__ = "0.16.1.dev202410100339"
+__version__ = "0.16.1.dev202410110340"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.16.1.dev202410100339.dist-info → keras_hub_nightly-0.16.1.dev202410110340.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: keras-hub-nightly
-Version: 0.16.1.dev202410100339
+Version: 0.16.1.dev202410110340
 Summary: Industry-strength Natural Language Processing extensions for Keras.
 Home-page: https://github.com/keras-team/keras-hub
 Author: Keras team

{keras_hub_nightly-0.16.1.dev202410100339.dist-info → keras_hub_nightly-0.16.1.dev202410110340.dist-info}/RECORD

@@ -1,15 +1,15 @@
 keras_hub/__init__.py,sha256=QGdXyHgYt6cMUAP1ebxwc6oR86dE0dkMxNy2eOCQtFo,855
 keras_hub/api/__init__.py,sha256=spMxsgqzjpeuC8rY4WP-2kAZ2qwwKRSbFwddXgUjqQE,524
 keras_hub/api/bounding_box/__init__.py,sha256=T8R_X7BPm0et1xaZq8565uJmid7dylsSFSj4V-rGuFQ,1097
-keras_hub/api/layers/__init__.py,sha256=jQbYVdHrjC0PnJVa3myfJbAmRef9KGwrTgPDaTsWPJw,2439
+keras_hub/api/layers/__init__.py,sha256=NpuVqxRCKbnvuW1QjZiSy7vHsI9Ej95ZCnNp4Lm5Hj8,2515
 keras_hub/api/metrics/__init__.py,sha256=So8Ec-lOcTzn_UUMmAdzDm8RKkPu2dbRUm2px8gpUEI,381
-keras_hub/api/models/__init__.py,sha256=Z3GLSmrvds7-XDtkcB9gq_sNZ5zJ_3Rwl_xoZ-dTX14,14786
+keras_hub/api/models/__init__.py,sha256=9-1qQyoGODam73Vlo49EI2fzTwzwgdlCshlTMRrFEqg,14884
 keras_hub/api/samplers/__init__.py,sha256=n-_SEXxr2LNUzK2FqVFN7alsrkx1P_HOVTeLZKeGCdE,730
 keras_hub/api/tokenizers/__init__.py,sha256=_f-r_cyUM2fjBB7iO84ThOdqqsAxHNIewJ2EBDlM0cA,2524
 keras_hub/api/utils/__init__.py,sha256=Gp1E6gG-RtKQS3PBEQEOz9PQvXkXaJ0ySGMqZ7myN7A,215
 keras_hub/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/api_export.py,sha256=9pQZK27JObxWZ96QPLBp1OBsjWigh1iuV6RglPGMRk0,1499
-keras_hub/src/version_utils.py,sha256=ta3Ul-3W63rzG1EOq-JStAvR39K8vw23TKBmNmLl2MQ,222
+keras_hub/src/version_utils.py,sha256=9zvnHnaU7MGFEG0NsmxQpWEG3qxlcuPtyLNw-vhmajA,222
 keras_hub/src/bounding_box/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/bounding_box/converters.py,sha256=a5po8DBm87oz2EXfi-0uEZHCMlCJPIb4-MaZIdYx3Dg,17865
 keras_hub/src/bounding_box/formats.py,sha256=YmskOz2BOSat7NaE__J9VfpSNGPJJR0znSzA4lp8MMI,3868
@@ -295,9 +295,9 @@ keras_hub/src/models/t5/t5_transformer_layer.py,sha256=uDeP84F1x7xJxki5iKe12Zn6e
 keras_hub/src/models/vae/__init__.py,sha256=i3UaSW4IJf76O7lSPE1dyxOVjuHx8iAYKivqvUbDHOw,62
 keras_hub/src/models/vae/vae_backbone.py,sha256=aYf1sGteFJ7FyR3X8Ek6QBjAT5GjRtQTK2jXhYVJeM4,6671
 keras_hub/src/models/vae/vae_layers.py,sha256=N83CYM1zgbl1EIjAOs3cFCkJEdxvbXkgM9ghKyljFAg,27752
-keras_hub/src/models/vgg/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-keras_hub/src/models/vgg/vgg_backbone.py,sha256=QnEDKn5n9bA9p3nvt5fBHnAssvnLxR0qv-oB372Ts0U,3702
-keras_hub/src/models/vgg/vgg_image_classifier.py,sha256=Dtq_HIJP6fHe8m7ZVLVn8IbHEsVMFWLvWMmn8TU1ntw,6600
+keras_hub/src/models/vgg/__init__.py,sha256=1ydFmkTOix2kOnDHie3srD4XD0dQ_7iR8OYbJzBM_YM,62
+keras_hub/src/models/vgg/vgg_backbone.py,sha256=qes1AsKwBDI7eQ3aC1uRievMkVNGXM9TNhtKLb9eZiU,3697
+keras_hub/src/models/vgg/vgg_image_classifier.py,sha256=bl6XM7l9fnOTGFreqOO3Z1jreusjhA4l7G0xjimfUKA,7829
 keras_hub/src/models/vit_det/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/models/vit_det/vit_det_backbone.py,sha256=GzwHXAfttExqDaGU4R2LAvng1gzjuvO3HMqUPwNUy9g,7656
 keras_hub/src/models/vit_det/vit_layers.py,sha256=oCKeUw5ckyUAGvmFPuxIiIAqgmC3uqh85LfZcgyh964,19852
@@ -353,7 +353,8 @@ keras_hub/src/utils/imagenet/imagenet_utils.py,sha256=MvIvv1WJo51ZXBxy4S7t_DsN3Z
 keras_hub/src/utils/timm/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/utils/timm/convert_densenet.py,sha256=V-GRjWuDnlh3b1EMxqahwZ3GMwSgOa3v0HOfb2ZZ-d0,3342
 keras_hub/src/utils/timm/convert_resnet.py,sha256=ee8eTml0ffJKE8avzGoLFcpjPF63DsvoIUArAGa8Ngg,5832
-keras_hub/src/utils/timm/preset_loader.py,sha256=CW-yNXvp3IExK3xuHKjYqbLdBVUST2kSsLmWxSs0i5c,2968
+keras_hub/src/utils/timm/convert_vgg.py,sha256=MT5jGnLrzenPpe66Af_Lp1IdR9KGtsSrcmn6_UPqHvQ,2419
+keras_hub/src/utils/timm/preset_loader.py,sha256=2GJI2YeKGVovtDqc930uGta12yiyuCL9YrsTyGhqt9Y,3094
 keras_hub/src/utils/transformers/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/utils/transformers/convert_albert.py,sha256=VdKclZpCxtDWq3UbUUQZf4fR9DJK_JYZ73B4O_G9skg,7695
 keras_hub/src/utils/transformers/convert_bart.py,sha256=Tk4h9Md9rwN5wjQbGIVrC7qzDpF8kI8qm-FKL8HlUok,14411
@@ -366,7 +367,7 @@ keras_hub/src/utils/transformers/convert_mistral.py,sha256=kVhN9h1ZFVhwkNW8p3wnS
 keras_hub/src/utils/transformers/convert_pali_gemma.py,sha256=B1leeDw96Yvu81hYumf66hIid07k5NLqoeWAJgPnaLs,10649
 keras_hub/src/utils/transformers/preset_loader.py,sha256=GS44hZUuGQCtzsyn8z44ZpHdftd3DFemwV2hx2bQa-U,2738
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=rPK-Uw1CG0DX0d_UAD-r2cG9fw8GI8bvAlrcXfQ9g4c,3323
-keras_hub_nightly-0.16.1.dev202410100339.dist-info/METADATA,sha256=Q1QCPvLIZpKSgKKhzvHH4Qng8vWrQihbFRVWk85Pjqs,7458
-keras_hub_nightly-0.16.1.dev202410100339.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-keras_hub_nightly-0.16.1.dev202410100339.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.16.1.dev202410100339.dist-info/RECORD,,
+keras_hub_nightly-0.16.1.dev202410110340.dist-info/METADATA,sha256=8UB_SWjRF9_PRraQNKuRN13SSTXmUCEwQPFGo0nU2Jk,7458
+keras_hub_nightly-0.16.1.dev202410110340.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
+keras_hub_nightly-0.16.1.dev202410110340.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.16.1.dev202410110340.dist-info/RECORD,,