Implementation of Dense and Convolutional Autoencoder in TensorFlow.
Different models can be found in the autoencoder.py file. New models can be added to the same file by following the format defined for other models.
Use trainer_ae.py for training with data loading mechanism implemented in Numpy and trainer_ae_queues.py for using the TensorFlow Queueing mechanism for input data loading.
TensorFlow Dataset API has been utilized for data reading in the latest version of the trainer i.e. trainer_ae_latest.py.
To start training of convolutional auto-encoder from scratch along with tensorboard visualization, use the following command:
python ./trainer_ae_latest.py -t -s --tensorboardVisualization --convolutionalAutoencoder --batchSize 50 --logsDir ./logs/ --modelDir ./model/
Similarly, for testing along with computation of the output image files, use the command:
python ./trainer_ae_latest.py -c --computeOutputImages --imagesOutputDirectory ./AutoEnc-output/
Author: Shoaib Ahmed Siddiqui
Email: 12bscsssiddiqui@seecs.edu.pk