AstroSegTrack

• This is the project supporting Joycelyn's master thesis
• We're segmenting and tracking super bubble using the magnetohydrodynamic simulation dataset provided by Alex

Data Preperation

Data/hdf5topng.py

• Converting astronomy h5df data type to png images
• Input: data root directory, output root path
• Output: All the sliced images to the designated folder, one per timestemp
• The output images are in grayscale each with size (1000, 1000)

Data/gt_construct.py

• Building the ground truth dataset for the Astro segmentation and tracking model

Model

Backbone

• I'm thinking adopting the latest SOTA video object segmentaion model - XMem++
• However, it's still heavily based on XMem, so we'll retrain based on the original XMem post

---

ToDO