Visual geo-localization seeks to enable autonomous aircraft positioning in GNSS-denied environments through large-scale image retrieval. Most existing methods, however, primarily rely on visible light images and are thus inherently sensitive to illumination changes and adverse weather conditions. Near-infrared (NIR) images offer superior environmental robustness but lack well-established benchmarks and specialized retrieval algorithms. To bridge this gap, we introduce NIR-cities, the first urban-scale benchmark for NIR aerial-to-satellite geo-localization. Unlike previous datasets, it features unaligned queries and continuous spatial coverage, simulating nadir-view aerial queries. Building on this, we propose a novel dual-context joint representation (DCJR) network specifically designed for NIR images. DCJR decouples the image representation into salient and environmental contexts via saliency-guided patch encoding (SPE). A dual-context enhancement (DCE) module then adaptively fuses these dual cues to address local ambiguities. Finally, region-prototype aggregation (RPA) distills the features through learnable prototypes, resulting in a discriminative and compact global descriptor. Extensive experiments demonstrate that DCJR significantly outperforms state-of-the-art methods, attaining a remarkable R@1 of 73.44% while achieving a favorable accuracy-efficiency tradeoff.
Teng-Da Zhang, Yunzhou Zhang, Li Wang et al.· IEEE Transactions on Geoscie...· 0 citations
Generative retrieval (GR) is a promising paradigm for industrial search advertising, yet its deployment is constrained by strict relevance and latency requirements. Existing systems cascade GR with an independent relevance model, decoupling the generative likelihood objective from query-ad relevance discrimination, which compromises effectiveness and increases serving costs. We propose a Unified Generative-Discriminative framework (UniGD) that integrates retrieval and relevance scoring within a single model. To mitigate gradient interference in joint optimization, UniGD introduces Conflict-Aware Gradient Enhancement (CAGE) to adaptively coordinate the two objectives. UniGD further designs a Codebook-Anchored Representation Module (CAM) that anchors item representations to frozen hierarchical codebooks distilled from a multimodal pretrained model, thereby endowing them with rich and generalizable semantic priors. For heterogeneous short-video, product, and live-stream ads, UniGD proposes Heterogeneous Ad-material Modeling (HAM), which captures cross-type semantic commonality over a shared backbone while preserving type-specific modeling capacity. Online AB tests on Kuaishou search advertising platform show that UniGD raises ad revenue by 5.78%, reduces inference latency by 33%, and improves discriminative relevance estimation. On NQ320K and MS300K, UniGD improves Recall@10 over the strongest reproduced GR baseline by 8.44% and 3.19%, respectively.
Shujie Ji, Yawei Kong, Yili Zhao et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
