REJEPA: A Novel Joint-Embedding Predictive Architecture for Efficient Remote Sensing Image Retrieval
This addresses the need for scalable and precise retrieval in remote sensing archives, offering a sensor-agnostic solution with incremental improvements over existing methods.
The paper tackles the problem of efficient content-based image retrieval for remote sensing by proposing REJEPA, a self-supervised framework that reduces computational complexity by 40-60% compared to pixel-reconstruction baselines and improves retrieval accuracy by 5.1-10.1% on multiple benchmarks.
The rapid expansion of remote sensing image archives demands the development of strong and efficient techniques for content-based image retrieval (RS-CBIR). This paper presents REJEPA (Retrieval with Joint-Embedding Predictive Architecture), an innovative self-supervised framework designed for unimodal RS-CBIR. REJEPA utilises spatially distributed context token encoding to forecast abstract representations of target tokens, effectively capturing high-level semantic features and eliminating unnecessary pixel-level details. In contrast to generative methods that focus on pixel reconstruction or contrastive techniques that depend on negative pairs, REJEPA functions within feature space, achieving a reduction in computational complexity of 40-60% when compared to pixel-reconstruction baselines like Masked Autoencoders (MAE). To guarantee strong and varied representations, REJEPA incorporates Variance-Invariance-Covariance Regularisation (VICReg), which prevents encoder collapse by promoting feature diversity and reducing redundancy. The method demonstrates an estimated enhancement in retrieval accuracy of 5.1% on BEN-14K (S1), 7.4% on BEN-14K (S2), 6.0% on FMoW-RGB, and 10.1% on FMoW-Sentinel compared to prominent SSL techniques, including CSMAE-SESD, Mask-VLM, SatMAE, ScaleMAE, and SatMAE++, on extensive RS benchmarks BEN-14K (multispectral and SAR data), FMoW-RGB and FMoW-Sentinel. Through effective generalisation across sensor modalities, REJEPA establishes itself as a sensor-agnostic benchmark for efficient, scalable, and precise RS-CBIR, addressing challenges like varying resolutions, high object density, and complex backgrounds with computational efficiency.