Beyond Prompt Degradation: Prototype-guided Dual-pool Prompting for Incremental Object Detection

Yaoteng Zhang, Qing Zhou, Junyu Gao, Qi Wang · 2026

Read Beyond Prompt Degradation: Prototype-guided Dual-pool Prompting for Incremental Object Detection by Yaoteng Zhang, Qing Zhou, Junyu Gao, Qi Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-02286.txt.

Description

Yaoteng Zhang, Qing Zhou, Junyu Gao, and Qi Wang improve step-by-step object detection in this 2026 paper. A detector must learn new object classes over time without forgetting old ones. Prompt-based methods are cheap and need no replay of old data. Yet linked prompts interfere, drift, and slowly degrade as tasks pile up.

Their framework, PDP, splits prompts into two pools with clear jobs. A shared pool holds general knowledge that transfers across tasks. A private pool holds features that mark each task's own classes. A second module builds class prototypes to label old objects that new data marks as background. That keeps training signals consistent across steps. Scores hit new highs on two major benchmarks.

Catalog record

CREATORYaoteng Zhang, Qing Zhou, Junyu Gao, Qi Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35802
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.02286

Browse the full library · About the Tower of Babel