# Catastrophic Forgetting in Sequential Thermal Anti-UAV Detection: The Role of Scale-Conditioned Gradient Imbalance

arXiv: https://arxiv.org/abs/2610.08315 · HTML: https://mlsanity.com/paper/2610.08315/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 17 of its 53 references that are in the map.
- First posted on arXiv 6 Oct 2026.

## Builds on (most cited first, 17 of 17)

- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Gradient Episodic Memory for Continual Learning](https://mlsanity.com/paper/1706.08840/) (2017)
- [Progressive Neural Networks](https://mlsanity.com/paper/1606.04671/) (2016)
- [Dark Experience for General Continual Learning: a Strong, Simple Baseline](https://mlsanity.com/paper/2004.07211/) (2020)
- [An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks](https://mlsanity.com/paper/1312.6211/) (2013)
- [On Tiny Episodic Memories in Continual Learning](https://mlsanity.com/paper/1902.10486/) (2019)
- [Gradient Starvation: A Learning Proclivity in Neural Networks](https://mlsanity.com/paper/2011.09468/) (2020)
- [Continual Learning with Low Rank Adaptation](https://mlsanity.com/paper/2311.17601/) (2023)
- [Evidential Detection and Tracking Collaboration: New Problem, Benchmark and Algorithm for Robust Anti-UAV System](https://mlsanity.com/paper/2306.15767/) (2023)
- [Unlocking the Power of Rehearsal in Continual Learning: A Theoretical Perspective](https://mlsanity.com/paper/2506.00205/) (2025)
- [CST Anti-UAV: A Thermal Infrared Benchmark for Tiny UAV Tracking in Complex Scenes](https://mlsanity.com/paper/2507.23473/) (2025)
- [Leveraging Foundation Models via Knowledge Distillation in Multi-Object Tracking: Distilling DINOv2 Features to FairMOT](https://mlsanity.com/paper/2407.18288/) (2024)
- [Uncertainty-Aware Multimodal Anti-UAV Detection via Evidential Fusion and Conflict-Discounted Belief Aggregation](https://mlsanity.com/paper/2608.29235/) (2026)
- [YOLOMG: Vision-based Drone-to-Drone Detection with Appearance and Pixel-Level Motion Fusion](https://mlsanity.com/paper/2503.07115/) (2025)
- [On The Relationship Between Continual Learning and Long-Tailed Recognition](https://mlsanity.com/paper/2306.13275/) (2023)
- [Evidential Deep Learning for Multi-Modal Anti-UAV Detection](https://mlsanity.com/paper/2609.01742/) (2026)
- [Evaluating the Robustness of Anti-UAV Detection under Controlled Fog Degradation: Fog-Aware Training and Clear-Sky Tradeoff](https://mlsanity.com/paper/2610.00141/) (2026)

## Abstract (arXiv)

Counter-UAV systems based on thermal infrared detection must stay accurate as operational datasets evolve, yet sequential fine-tuning causes catastrophic forgetting of prior tasks, a problem that remains insufficiently characterized in this domain. This continual-learning study measures the stability-plasticity trade-off in YOLOMG, a YOLOv5-based detector run as a single thermal-infrared stream with the motion channel disabled, trained sequentially across three anti-UAV benchmarks of rising scale difficulty: Anti-UAV-RGBT, Anti-UAV410, and CST Anti-UAV. Naive fine-tuning on CST yields a Forgetting Measure of -0.605 against the Stage 1 ceiling, corresponding to a 90% capability loss, with -0.572 occurring in Stage 3 alone. In contrast, knowledge distillation from a frozen teacher is associated with FM = -0.033 +/- 0.004 across three seeds, corresponding to 95% retention. Because no Stage 2 no-KD control is included, this result establishes retention under KD training rather than a causal KD effect. Per-stratum analysis shows large-target detection collapsing to near zero within the first epoch, despite an inter-stage cosine similarity of 0.987 over the gradient-updated weights, pointing to scale-conditioned gradient imbalance, rather than weight drift, as a candidate mechanism. Scale-Stratified Herding (SSH), a 300-exemplar buffer balanced across four UAV size strata, roughly halves the forgetting (FM = -0.605 to -0.311) and keeps large-target detection non-zero. An ablation attributes the gain primarily to scale stratification rather than herding: random-stratified replay performs at least as well (FM = -0.221 versus -0.311 for SSH). These replay results are single-seed and should therefore be treated as preliminary.