Task-Guided Multi-Annotation Triplet Learning for Remote Sensing Representations

TL;DR

This paper introduces a task-guided triplet loss method for remote sensing that adaptively selects informative triplets across tasks, improving multi-task representation learning without static weighting.

Contribution

It proposes a mutual-information based triplet selection strategy that enhances multi-task learning by dynamically choosing samples, eliminating the need for static loss weights.

Findings

Improved classification and regression performance on aerial wildlife dataset.

Task-aware triplet selection yields more effective shared representations.

Outperforms several static-weight triplet loss setups.

Abstract

Prior multi-task triplet loss methods relied on static weights to balance supervision between various types of annotation. However, static weighting requires tuning and does not account for how tasks interact when shaping a shared representation. To address this, the proposed task-guided multi-annotation triplet loss removes this dependency by selecting triplets through a mutual-information criteria that identifies triplets most informative across tasks. This strategy modifies which samples influence the representation rather than adjusting loss magnitudes. Experiments on an aerial wildlife dataset compare the proposed task-guided selection against several triplet loss setups for shaping a representation in an effective multi-task manner. The results show improved classification and regression performance and demonstrate that task-aware triplet selection produces a more effective shared representation for downstream tasks.