Video Scene Graph Generation (VidSGG) aims to capture dynamic relationships among entities by sequentially analyzing video frames and integrating visual and semantic information. However, VidSGG is challenged by significant biases that skew predictions. To mitigate these biases, we propose a VIsual and Semantic Awareness (VISA) framework for unbiased VidSGG. VISA addresses visual bias through memory-enhanced temporal integration that enhances object representations and concurrently reduces semantic bias by iteratively integrating object features with comprehensive semantic information derived from triplet relationships. This visual-semantics dual debiasing approach results in more unbiased representations of complex scene dynamics. Extensive experiments demonstrate the effectiveness of our method, where VISA outperforms existing unbiased VidSGG approaches by a substantial margin (e.g., +13.1% improvement in mR@20 and mR@50 for the SGCLS task under Semi Constraint).

Images captured under challenging lighting conditions suffer from both uneven illumination and flare artifacts (e.g., glare, streaks, and shimmer). Existing image enhancement methods mainly focus on either enhancing low-light regions or removing flares, but rarely address both issues simultaneously. When these methods are applied in sequence, they inevitably lead to over-enhancement and saturation in bright regions affected by flares or insufficient enhancement in low-light areas. In this article, we introduce a neural network model – IllumiNet, to enhance images captured under complex lighting conditions. Specifically, we propose a two-stage pixel-to-pixel generative model , that achieves both image flare removal and image light enhancement. Each stage is structured based on the U-Net model with the Mix Vision Transformer as the backbone, which is shared by the two stages to support domain knowledge interaction between the two tasks while helping to reduce model size. Additionally, we introduce a training strategy that combines knowledge distillation with simulated data generation, eliminating the need for real-paired datasets. Experiments across various benchmarks demonstrate that IllumiNet produces visually appealing enhanced images and exhibits robust generalization to diverse real-world scenarios.