Uemerson / masters-degree-in-computer-science-from-unesp

Master's Degree in Computer Science from UNESP

Paper(s)

Registro Não-Rígido de Imagens Médicas usando Block-Based Principal Component Analysis como Camada de Pooling

About

This repository stores the training scripts, submission, and other files used in the research entitled "Registro Não-Rígido de Imagens Médicas usando Block-Based Principal Component Analysis como Camada de Pooling" conducted as part of the master's degree program at São Paulo State University (UNESP).

Datasets

You can download the datasets used in this work from:

• 👉 Download IXI
• 👉 Download OASIS

How to train ?

The train scripts are inside OASIS and IXI folder, start with train_*

OASIS

The following scripts train using the OASIS dataset:

• train_oasis_vxm_2.py: Trains Voxelmorph without a pooling layer (convolution with stride).
• train_oasis_vxm_2_max.py: Trains Voxelmorph with a Max Pooling layer.
• train_oasis_vxm_2_bpca.py: Trains Voxelmorph with a BPCA (Block-Based Principal Component Analysis) pooling layer.
• train_oasis_vxm_2_bpca_revert.py: Trains Voxelmorph with a BPCA pooling layer and reverts BPCA on the unpooling layer.
• train_oasis_vxm_2_max_bpca.py: Trains Voxelmorph with a mix of the first 3 pooling layers using Max Pooling and the final layer using BPCA.
• train_oasis_vxm_2_bpca_max.py: Trains Voxelmorph with a mix of the first 3 pooling layers using BPCA and the final layer using Max Pooling.

IXI

The following scripts train using the IXI dataset:

• train_ixi_vxm_2.py: Trains Voxelmorph without a pooling layer (convolution with stride).
• train_ixi_vxm_2_max.py: Trains Voxelmorph with a Max Pooling layer.
• train_ixi_vxm_2_bpca.py: Trains Voxelmorph with a BPCA (Block-Based Principal Component Analysis) pooling layer.
• train_ixi_vxm_2_bpca_revert.py: Trains Voxelmorph with a BPCA pooling layer and reverts BPCA on the unpooling layer.
• train_ixi_vxm_2_max_bpca.py: Trains Voxelmorph with a mix of the first 3 pooling layers using Max Pooling and the final layer using BPCA.
• train_ixi_vxm_2_bpca_max.py: Trains Voxelmorph with a mix of the first 3 pooling layers using BPCA and the final layer using Max Pooling.

How to use tensorboard ?

After installing dependencies and activating environments using conda, you can:

$ tensorboard --logdir path-to-log/logs --port 6006

Observation

In some scripts, you will need to change the foler path, for example, /home/hinton/uemerson/ to your directory path.