Learning on the Fly - Presentation Website

This repository contains the source code and assets for the official project website of the paper "Learning on the Fly: Replay-Based Continual Object Perception for Indoor Drones".

Live Website: spacetime-vision-robotics-laboratory.github.io/learning-on-the-fly-cl

Paper Summary

Authors: Sebastian-Ion Nae, Mihai-Eugen Barbu, Sebastian Mocanu, Marius Leordeanu

Abstract:

Autonomous agents such as indoor drones must learn new object classes in real-time while limiting catastrophic forgetting, motivating Class-Incremental Learning (CIL). We introduce an indoor dataset of 14,400 frames capturing inter-drone and ground vehicle footage, annotated via a semi-automatic workflow with 98.6% first-pass labeling agreement. Using this dataset, we benchmark three replay-based CIL strategies: Experience Replay (ER), Maximally Interfered Retrieval (MIR), and Forgetting-Aware Replay (FAR), using YOLOv11n as a resource-efficient detector. Across strict memory budgets (5–10% replay), FAR yields the best performance, achieving an average accuracy of 82.96% (mAP₅₀₋₉₅) with 5% replay.

Resources

• Paper (PDF)
• arXiv
• Code
• Dataset

Local Development

Simply open index.html in a browser, or serve with any static file server:

python -m http.server 8000

Deployment

The site auto-deploys to GitHub Pages via the included workflow .github/workflows/github-pages.yml, which handles image compression and CSS/JS minification.

Citation

@article{nae2026learningflyreplaybasedcontinual,
    title   = {Learning on the Fly: Replay-Based Continual Object Perception for Indoor Drones},
    author  = {Nae, Sebastian-Ion and Barbu, Mihai-Eugen and Mocanu, Sebastian and Leordeanu, Marius},
    journal = {arXiv preprint arXiv:2602.13440},
    year    = {2026}
}