Reproduce “EO-Jev: Decision Model for Earth Observation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Mohamed Atef Abouelhadid, Diaa Addeen Abuhani, Muhammad Akhtar Munir, Fahad Shahbaz Khan, Salman Khan

EO-Jev adapts the 8.4-billion-parameter d3-flash model to make one-pass decisions across 30 settings from 23 public benchmarks. It scores predefined answers with probabilities, using low-rank adapters and EO training data. On 26 of 30 settings it improves over d3-flash; across 12 datasets entirely held out from adaptation, mean performance rises 8.0 percentage points, though some tasks remain difficult.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.