TensorFlow implementation of AlexNet(2012) by Alex Krizhevsky, with 3D convolutiional layers.
Network with a standart AlexNet architecture, but it has 3D instead 2D filters on each Conv and Pool layers.
To fit this model into memory, model training proccess can be refactored with @OpenAI's Gradient Checkpointing algorithm https://github.com/openai/gradient-checkpointing
-
Note: This model needs a lot of GPU memory. Training session will not starts on 1 GPU without additional data separation or code's parallelization (or https://github.com/openai/gradient-checkpointing for example).
-
Note2: This model doesn't pretend to be the
SilverBulletin 3D image recognition (like AlexNet was in 2D). It's just an example of 3D convolutional model in TensorFlow.