Welcome to the official repository of our paper "Scale-Invariant Features Adversarial Disentanglement for UAV Object Detection"!
In this paper, we propose enhancing the UAV-OD accuracy via Scale-Invariant Features Adversarial Disentanglement. Firstly, we designed a Scale-Invariant Features Disentanglement module through Adversarial Learning (SIFAD). Then we construct a multi-scene and multi-modal UAV-OD dataset. Experimental results demonstrated the superiority of our approach. This is just our preliminary exploration of disentangling scale-invariant features. More subtle and effective designs can be considered, leaving enough space for further development. Moreover, it is worth mentioning that we are committed to further enhancing the scope and scale of State-Air, expanding both the coverage and depth of our data.
In this repository, we will provide our code (SIFAD) and Dataset (State-Air).