feat(Architectures): Implemented Vision Transformer Architecture

aucmedi/neural_network/architectures/image/__init__.py

@@ -64,6 +64,9 @@
 # Xception
 from aucmedi.neural_network.architectures.image.xception import Architecture_Xception
 # Vision Transformer (ViT)
+from aucmedi.neural_network.architectures.image.vit_b16 import Architecture_ViT_B16
+from aucmedi.neural_network.architectures.image.vit_b32 import Architecture_ViT_B32
+from aucmedi.neural_network.architectures.image.vit_l16 import Architecture_ViT_L16
 from aucmedi.neural_network.architectures.image.vit_l32 import Architecture_ViT_L32
 
 #-----------------------------------------------------#
@@ -100,6 +103,9 @@
     "VGG16": Architecture_VGG16,
     "VGG19": Architecture_VGG19,
     "Xception": Architecture_Xception,
+    "ViT_B16": Architecture_ViT_B16,
+    "ViT_B32": Architecture_ViT_B32,
+    "ViT_L16": Architecture_ViT_L16,
     "ViT_L32": Architecture_ViT_L32,
 }
 """ Dictionary of implemented 2D Architectures Methods in AUCMEDI.
@@ -182,6 +188,9 @@
     "VGG16": "caffe",
     "VGG19": "caffe",
     "Xception": "tf",
+    "ViT_B16": "tf",
+    "ViT_B32": "tf",
+    "ViT_L16": "tf",
     "ViT_L32": "tf",
 }
 """ Dictionary of recommended [Standardize][aucmedi.data_processing.subfunctions.standardize] techniques for 2D Architectures Methods in AUCMEDI.

aucmedi/neural_network/architectures/image/vit_b16.py

@@ -0,0 +1,123 @@
+#==============================================================================#
+#  Author:       Dominik Müller                                                #
+#  Copyright:    2022 IT-Infrastructure for Translational Medical Research,    #
+#                University of Augsburg                                        #
+#                                                                              #
+#  This program is free software: you can redistribute it and/or modify        #
+#  it under the terms of the GNU General Public License as published by        #
+#  the Free Software Foundation, either version 3 of the License, or           #
+#  (at your option) any later version.                                         #
+#                                                                              #
+#  This program is distributed in the hope that it will be useful,             #
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of              #
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
+#  GNU General Public License for more details.                                #
+#                                                                              #
+#  You should have received a copy of the GNU General Public License           #
+#  along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
+#==============================================================================#
+#-----------------------------------------------------#
+#                    Documentation                    #
+#-----------------------------------------------------#
+""" The classification variant of the Vision Transformer (ViT) version B16 architecture.
+
+!!! warning
+    The ViT architectures only work for RGB encoding (channel size = 3).
+
+| Architecture Variable    | Value                      |
+| ------------------------ | -------------------------- |
+| Key in architecture_dict | "2D.ViT_B16"               |
+| Input_shape              | (224, 224)                 |
+| Standardization          | "tf"                       |
+
+???+ abstract "Reference - Implementation"
+    Fausto Morales; https://github.com/faustomorales <br>
+    https://github.com/faustomorales/vit-keras <br>
+
+    Vo Van Tu; https://github.com/tuvovan <br>
+    https://github.com/tuvovan/Vision_Transformer_Keras <br>
+
+    Original: Google Research <br>
+    https://github.com/google-research/vision_transformer <br>
+
+???+ abstract "Reference - Publication"
+    ```
+    @article{dosovitskiy2020vit,
+      title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
+      author={Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and  Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil},
+      journal={ICLR},
+      year={2021}
+    }
+
+    @article{tolstikhin2021mixer,
+      title={MLP-Mixer: An all-MLP Architecture for Vision},
+      author={Tolstikhin, Ilya and Houlsby, Neil and Kolesnikov, Alexander and Beyer, Lucas and Zhai, Xiaohua and Unterthiner, Thomas and Yung, Jessica and Steiner, Andreas and Keysers, Daniel and Uszkoreit, Jakob and Lucic, Mario and Dosovitskiy, Alexey},
+      journal={arXiv preprint arXiv:2105.01601},
+      year={2021}
+    }
+
+    @article{steiner2021augreg,
+      title={How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers},
+      author={Steiner, Andreas and Kolesnikov, Alexander and and Zhai, Xiaohua and Wightman, Ross and Uszkoreit, Jakob and Beyer, Lucas},
+      journal={arXiv preprint arXiv:2106.10270},
+      year={2021}
+    }
+
+    @article{chen2021outperform,
+      title={When Vision Transformers Outperform ResNets without Pretraining or Strong Data Augmentations},
+      author={Chen, Xiangning and Hsieh, Cho-Jui and Gong, Boqing},
+      journal={arXiv preprint arXiv:2106.01548},
+      year={2021},
+    }
+
+    @article{zhai2022lit,
+      title={LiT: Zero-Shot Transfer with Locked-image Text Tuning},
+      author={Zhai, Xiaohua and Wang, Xiao and Mustafa, Basil and Steiner, Andreas and Keysers, Daniel and Kolesnikov, Alexander and Beyer, Lucas},
+      journal={CVPR},
+      year={2022}
+    }
+    ```
+"""
+#-----------------------------------------------------#
+#                   Library imports                   #
+#-----------------------------------------------------#
+# External libraries
+from vit_keras import vit
+# Internal libraries
+from aucmedi.neural_network.architectures import Architecture_Base
+
+#-----------------------------------------------------#
+#     Architecture class: Vision Transformer (ViT)    #
+#-----------------------------------------------------#
+class Architecture_ViT_B16(Architecture_Base):
+    #---------------------------------------------#
+    #                Initialization               #
+    #---------------------------------------------#
+    def __init__(self, classification_head, channels, input_shape=(384, 384),
+                 pretrained_weights=False):
+        self.classifier = classification_head
+        self.input = input_shape + (channels,)
+        self.pretrained_weights = pretrained_weights
+
+    #---------------------------------------------#
+    #                Create Model                 #
+    #---------------------------------------------#
+    def create_model(self):
+        # Get pretrained image weights from imagenet if desired
+        if self.pretrained_weights : pretrained = True
+        else : pretrained = False
+
+        # Obtain ViT B16 as base model
+        base_model = vit.vit_b16(image_size=self.input[:-1],
+                                 classes=self.classifier.n_labels,
+                                 include_top=False,
+                                 pretrained=pretrained,
+                                 pretrained_top=False)
+        top_model = base_model.output
+
+        # Add classification head
+        model = self.classifier.build(model_input=base_model.input,
+                                      model_output=top_model)
+
+        # Return created model
+        return model

aucmedi/neural_network/architectures/image/vit_b32.py

@@ -0,0 +1,123 @@
+#==============================================================================#
+#  Author:       Dominik Müller                                                #
+#  Copyright:    2022 IT-Infrastructure for Translational Medical Research,    #
+#                University of Augsburg                                        #
+#                                                                              #
+#  This program is free software: you can redistribute it and/or modify        #
+#  it under the terms of the GNU General Public License as published by        #
+#  the Free Software Foundation, either version 3 of the License, or           #
+#  (at your option) any later version.                                         #
+#                                                                              #
+#  This program is distributed in the hope that it will be useful,             #
+#  but WITHOUT ANY WARRANTY; without even the implied warranty of              #
+#  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               #
+#  GNU General Public License for more details.                                #
+#                                                                              #
+#  You should have received a copy of the GNU General Public License           #
+#  along with this program.  If not, see <http://www.gnu.org/licenses/>.       #
+#==============================================================================#
+#-----------------------------------------------------#
+#                    Documentation                    #
+#-----------------------------------------------------#
+""" The classification variant of the Vision Transformer (ViT) version B32 architecture.
+
+!!! warning
+    The ViT architectures only work for RGB encoding (channel size = 3).
+
+| Architecture Variable    | Value                      |
+| ------------------------ | -------------------------- |
+| Key in architecture_dict | "2D.ViT_B32"               |
+| Input_shape              | (224, 224)                 |
+| Standardization          | "tf"                       |
+
+???+ abstract "Reference - Implementation"
+    Fausto Morales; https://github.com/faustomorales <br>
+    https://github.com/faustomorales/vit-keras <br>
+
+    Vo Van Tu; https://github.com/tuvovan <br>
+    https://github.com/tuvovan/Vision_Transformer_Keras <br>
+
+    Original: Google Research <br>
+    https://github.com/google-research/vision_transformer <br>
+
+???+ abstract "Reference - Publication"
+    ```
+    @article{dosovitskiy2020vit,
+      title={An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
+      author={Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and  Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil},
+      journal={ICLR},
+      year={2021}
+    }
+
+    @article{tolstikhin2021mixer,
+      title={MLP-Mixer: An all-MLP Architecture for Vision},
+      author={Tolstikhin, Ilya and Houlsby, Neil and Kolesnikov, Alexander and Beyer, Lucas and Zhai, Xiaohua and Unterthiner, Thomas and Yung, Jessica and Steiner, Andreas and Keysers, Daniel and Uszkoreit, Jakob and Lucic, Mario and Dosovitskiy, Alexey},
+      journal={arXiv preprint arXiv:2105.01601},
+      year={2021}
+    }
+
+    @article{steiner2021augreg,
+      title={How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers},
+      author={Steiner, Andreas and Kolesnikov, Alexander and and Zhai, Xiaohua and Wightman, Ross and Uszkoreit, Jakob and Beyer, Lucas},
+      journal={arXiv preprint arXiv:2106.10270},
+      year={2021}
+    }
+
+    @article{chen2021outperform,
+      title={When Vision Transformers Outperform ResNets without Pretraining or Strong Data Augmentations},
+      author={Chen, Xiangning and Hsieh, Cho-Jui and Gong, Boqing},
+      journal={arXiv preprint arXiv:2106.01548},
+      year={2021},
+    }
+
+    @article{zhai2022lit,
+      title={LiT: Zero-Shot Transfer with Locked-image Text Tuning},
+      author={Zhai, Xiaohua and Wang, Xiao and Mustafa, Basil and Steiner, Andreas and Keysers, Daniel and Kolesnikov, Alexander and Beyer, Lucas},
+      journal={CVPR},
+      year={2022}
+    }
+    ```
+"""
+#-----------------------------------------------------#
+#                   Library imports                   #
+#-----------------------------------------------------#
+# External libraries
+from vit_keras import vit
+# Internal libraries
+from aucmedi.neural_network.architectures import Architecture_Base
+
+#-----------------------------------------------------#
+#     Architecture class: Vision Transformer (ViT)    #
+#-----------------------------------------------------#
+class Architecture_ViT_B32(Architecture_Base):
+    #---------------------------------------------#
+    #                Initialization               #
+    #---------------------------------------------#
+    def __init__(self, classification_head, channels, input_shape=(384, 384),
+                 pretrained_weights=False):
+        self.classifier = classification_head
+        self.input = input_shape + (channels,)
+        self.pretrained_weights = pretrained_weights
+
+    #---------------------------------------------#
+    #                Create Model                 #
+    #---------------------------------------------#
+    def create_model(self):
+        # Get pretrained image weights from imagenet if desired
+        if self.pretrained_weights : pretrained = True
+        else : pretrained = False
+
+        # Obtain ViT B32 as base model
+        base_model = vit.vit_b32(image_size=self.input[:-1],
+                                 classes=self.classifier.n_labels,
+                                 include_top=False,
+                                 pretrained=pretrained,
+                                 pretrained_top=False)
+        top_model = base_model.output
+
+        # Add classification head
+        model = self.classifier.build(model_input=base_model.input,
+                                      model_output=top_model)
+
+        # Return created model
+        return model