jlouvis / deep-learning

Deep-Learning

Synopsis

Project Title: Automating the Segmentation of X-ray Images with Deep Neural Networks

Team Member: Shuli Sun S232245, Søren Blatt Bendtsen s164521, Pantelis Apostolidis s230697, Ioannis Louvis s222556

Motivation & Background: It is a time-consuming and error-prone process to segment the result images from the X-ray physics experiments. So, we want to build a deep-learning model to do it.

Milestones:

Week 1:

Read through references (Image segmentation, UNet, reports by supervisor)

Read in the data to Python / Pytorch

Visualize a few images

See if any type of data preparation / cleaning is needed

Week 2:

Have access to GPU (using the HPC guide uploaded by the TA’s)

Start to build a simple model to train the data (Study for Networks like VGGNET, UNet)

Encoder, Decoder

Week 3 – 4:

Finetune model and hyperparameters

Try different techniques

Neural Network

Week 5:

Work on Poster Session

Write report

Week 6 – 7:

Finetune model

Finish report

References:

De Angelis, S. et al. Three-dimensional characterization of nickel coarsening in solid oxide cells via ex-situ ptychographic nano-tomography. Journal of Power Sources 383, 72–79 (2018).

De Angelis, S. et al. Ex-situ tracking solid oxide cell electrode microstructural evolution in a redox cycle by high resolution ptychographic nanotomography. Journal of Power Sources 360, 520–527 (2017).