30 June 2025 in IEEE Access
Person Re-identification (Re-ID) aims at accurately querying pedestrians across multiple non-overlapping cameras system, playing an essential role in computer vision applications. While CNN-based methods leverage feature aggregation and attention mechanisms to achieve competitive performance, their limitations in capturing long-range dependencies have motivated the research of transformer-based architectures, which excel in the evolution of person Re-ID research. Despite leveraging pretrained vision-language models to facilitate visual encoder training, current methods suffer from fragmented optimization procedures, complexity increasing, and inadequate semantic guidance from textual prompts. To address these limitations, we propose an end-to-end method named Text-Guided Fusion Transformer (TGFT). TGFT applies the fixed but semantically enriched texts to guide visual training with a novel Gated Cross Attention (GCA) fusion module. The GCA module adaptively modulates the integration of textual information and visual features, effectively enhancing cross-modal feature alignment and semantic consistency. Without significantly increasing computational complexity or parameter count, TGFT achieves superior performance, demonstrating strong effectiveness and scalability on multiple major Re-ID benchmarks.
23 September 2025 in IEEE Access
Remote sensing change detection (RSCD) must remain reliable under illumination, seasonal, and viewpoint variability while handling objects that range from narrow roads to block-scale complexes. We present Efficient Spectral Inception Former Change Detector (ESIFCD), an efficient dual-temporal network that couples domain-aligned differencing with multiple cross-resolution exchange and lightweight global context encoding. ESIFCD introduces: (i) an Adaptive Difference Fusion Unit (ADFU) that fuses absolute and learned differences over spatial, spectral, and texture features; (ii) a Cross-Resolution Aggregation Module (CRAM) that repeatedly exchanges evidence among four scales to mitigate scale disagreement; (iii) a Patch-Aware Context Encoder (PACE) that models long-range dependencies via tokenized patches with modest cost; and (iv) a fast GLCM-Stats Encoder that supplies texture semantics (ASM/Contrast/Correlation) without analytical histograms. A re-parameterized large-kernel block (RLKB) sharpens local detail, and a late MetaCNNGate learns pixel-wise weights over one primary and four auxiliary heads to consolidate their complementary expertise (small vs. large structures). Across six public benchmarks, ESIFCD attains the best F1 on each dataset: 93.19% (LEVIR-CD), 87.89% (LEVIR-CD+), 70.10% (S2Looking), 95.95% (CDD), 83.48% (SYSU-CD), and 95.30% (WHU-CD). The model is efficient and fast (12.33 GFLOPs; 82.07 images/s at 2562), and with automatic mixed precision reaches over 165 images/s while maintaining accuracy, surpassing recent strong baselines under the same setting. Ablations confirm that ADFU improves robustness to radiometric and registration noise, CRAM preserves thin structures and prevents fragmentation on large objects, PACE improves cluttered scenes, and MetaCNNGate consistently lifts F1 on all benchmarks.