ΕΕΛΛΑΚ - Λίστες Ταχυδρομείου

Clarification on proposal direction - Open-Source AI Framework for Thermal Satellite Payload Data Analysis

  • Subject: Clarification on proposal direction - Open-Source AI Framework for Thermal Satellite Payload Data Analysis
  • From: Nikitas Karachalios <karachaliosnikitas [ at ] gmail [ dot ] com>
  • Date: Tue, 17 Mar 2026 17:40:04 +0200
Dear Christos, Alexis and Simon,

I hope you are all doing well. I have been studying the TIRAuxCloud paper
and repository carefully, and I have already researched the uncertainty
quantification/explainability direction in depth, especially single-pass
uncertainty, Grad-CAM++, and DeepSHAP. I have also scanned part of the
paper’s references to better understand the technical directions around the
project. My proposal is already centered on uncertainty
quantification/explainability as the core direction. For the second part, I
am now trying to choose one realistic beyond-supervised direction from the
future-work points discussed in the paper. Right now I see two concrete
possibilities:

1. a direction closer to the current repository, by benchmarking and
extending the existing cloudy/clear paired models already present in the
codebase, such as Siamese and bam-cd
2. one focused self-supervised learning study, for example a SatMAE-style
masked autoencoder pretraining approach, motivated by the paper’s
future-work

I also noticed the paper mentions other future directions such as
foundation-model evaluation, lightweight in-orbit architectures, and while
scanning the codebase I also came across DINOv2- and SAM-related
components. Before fixing this part of my proposal, I wanted to ask whether
there is any current preference among these beyond-supervised directions.
In particular, would you prefer a secondary direction that is closer to the
current codebase, such as the existing change-aware cloudy/clear models, or
should I propose the direction I currently feel more confident I can
execute well, namely a focused self-supervised study? If the foundation
model direction is currently considered more relevant than these two, I
would also appreciate knowing that. I ask this because I want the proposal
to stay technically clear and realistic in scope. In principle, I think
both of the two concrete directions above could be valuable, and if time
allows I would be happy to include both in some form. At the same time,
although I have tried to prepare for the project as thoroughly as possible,
I have not yet worked this deeply in a project of this kind, so I am not
fully confident in deciding the most realistic scope by myself. That is why
I would really appreciate your guidance on whether one of these directions
is currently preferred for the repository.

Best regards,
Nikitas
----
Λαμβάνετε αυτό το μήνυμα απο την λίστα: Λίστα αλληλογραφίας και συζητήσεων που απευθύνεται σε φοιτητές developers \& mentors έργων του Google Summer of Code - A discussion list for student developers and mentors of Google Summer of Code projects.,
https://lists.ellak.gr/gsoc-developers/listinfo.html
Μπορείτε να απεγγραφείτε από τη λίστα στέλνοντας κενό μήνυμα ηλ. ταχυδρομείου στη διεύθυνση <gsoc-developers+unsubscribe [ at ] ellak [ dot ] gr>.

πλοήγηση μηνυμάτων