Reproduce “Removal then Selection: A Coarse-to-Fine Fusion Perspective for RGB-Infrared Object Detection” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tianyi Zhao, Maoxun Yuan, Feng Jiang, Nan Wang, Xingxing Wei

Researchers from Beihang University developed a coarse-to-fine fusion strategy for RGB-Infrared object detection, inspired by human selective attention, that first removes redundant information and then dynamically selects features. This approach achieved state-of-the-art performance on the KAIST, FLIR, and LLVIP datasets, demonstrating improved accuracy, particularly for small or occluded objects in challenging conditions.

Key insights

  • Inspired by cognitive psychology's Attenuation Theory, explicitly removing redundant or interfering information from modalities before precise feature selection leads to more effective multi-modal fusion.
  • Operating in the frequency domain allows for coarse but effective removal of irrelevant and noisy spectral components within individual RGB and Infrared images.
  • Dynamically selecting and combining features across multiple scales, using a mixture of scale-aware experts, significantly enhances the discriminative power of fused representations, especially for varying object sizes.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.