marinadunn / europa-chaos-ml

Automated instance segmentation for ice blocks within Europa's chaos terrain using corrected NASA Galileo images.

Hi, I'm Marina! 👋

Pronouns: she/her/hers

marinadunn.github.io

✨About Me✨

🔭 My research interests include: Data Science and Deep Learning for Astrophysics research

👩‍💻 Starting in September 2024, I will begin my role as a Junior Scientist (Astrophysics PhD Candidate) at the Instituto de Astrofísica de Canarias (IAC) in Tenerife, Spain!

👩‍🎓 I recently graduated in June 2023 with my M.S. in Engineering: Data Science from the University of Riverside, California

🌌 I received my Bachelor's in Astronomy from the University of Arizona in 2018

🚀 I've previously worked at NASA Goddard Space Flight Center, NASA Langley Research Center, Lawrence Livermore National Laboratory, Apple, and the University of Arizona

🧠 I am proudly neurodivergent!

📍 Hometown: Nashville, TN, USA

✉️ Get in touch: [email protected] (same as [email protected]) or [email protected]

🛠️ Skills

👩‍💻 Languages

📈 Artificial Intelligence/Machine Learning/Deep Learning/Data Science

⚙️ Version Control

📚 Libraries

💻 IDEs/Editors

europa-chaos-ml's Issues

Question About Hyperparameter Search

Hi Ahmed,

I am working on consolidating the code and writing documentation for reproducibility. I had a question about the hyperparameter search: for the optimal parameters we listed (learning rate of 0.001, batch size of 1, window crop size of 250x250, stride of 64, Adam optimizer, training with 3 trainable backbone layers, and training duration of 15 epochs), to confirm- was this with the average F1 score objective function (avg_f1_objective) or something else? And what was the min_sheet_area value used for that? Thanks!

Add current model weights

Please add the completed trained PyTorch model to a new directory models in main repo for the model referenced in the NeurIPS paper, and associated weights file to models/model_weights

Confirm run status of cross_val.py

Confirm whether src/cross_val.py is able to run as-is, otherwise modify/update repo to current file for running LOOCV from start to finish as described in NeurIPS paper.