Search papers, labs, and topics across Lattice.
This paper introduces FressDet, a fully rotation-equivariant framework for multispectral object detection that addresses limitations in existing detectors related to spectral processing and spatial cue reliability. By integrating components such as Spectral Implicit Warp for query-based spectral resampling and Rotation-Equivariant Consistency Weighting for adaptive fusion of spectral and spatial cues, FressDet enhances the robustness of object detection under arbitrary rotations. The approach achieves state-of-the-art performance with 93% fewer parameters across three public benchmarks, highlighting its efficiency and effectiveness in real-world applications.
FressDet achieves state-of-the-art multispectral object detection while using 93% fewer parameters, revolutionizing efficiency in the field.
Existing multispectral detectors are limited by discrete spectral processing, a scale-dependent shift in the relative reliability of spectral and spatial cues across pyramid levels, and the lack of explicit rotation-equivariant geometric priors for arbitrarily oriented objects. To tackle these limitations, we propose FressDet, a fully rotation-equivariant spectral-spatial learning framework for multispectral object detection, capable of capturing the continuous, ordered nature of spectral structure and enabling reliable spectral-spatial fusion across pyramid levels under arbitrary in-plane rotations. FressDet integrates three complementary components. Spectral Implicit Warp (SpeIW) enables query-based spectral resampling via a coordinate-conditioned implicit field, yielding a monotone, order-preserving warp. Rotation-Equivariant Consistency Weighting (ReCoW) adaptively fuses spectral and spatial branches based on branch reliability, reinforcing informative cues while suppressing noise across pyramid levels. The oriented-aware head exploits group-indexed features to stably predict oriented objects without parameter replication. Taken together, FressDet learns more discriminative and robust spectral-spatial representations even under rotational perturbations. By achieving state-of-the-art performance with 93% fewer parameters on three public benchmarks, FressDet demonstrates its effectiveness and generalizability.
