핵심 개념
Efficiently learning to generate darkness clue prompts for robust UAV tracking at night.
초록
Existing nighttime UAV trackers follow an "Enhance-then-Track" architecture, which fails to build an end-to-end trainable vision system. To address this, DCPT proposes a novel architecture that achieves robust UAV tracking at night by efficiently learning to generate darkness clue prompts. The proposed method directly encodes anti-dark capabilities into prompts using a darkness clue prompter (DCP). By injecting these learned visual prompts into a daytime tracker with fixed parameters across transformer layers, DCPT enables adaptive fusion between prompts and the base model. Extensive experiments show state-of-the-art performance for DCPT on multiple dark scenario benchmarks. The unified end-to-end learning of enhancement and tracking in DCPT enables a more trainable system without extra modules.
통계
Code available at https://github.com/bearyi26/DCPT.
DCPT boosts the base tracker by 4.9% success score with 3.0M prompting parameters.