Predicting pedestrian trajectories in crowded and dynamic scenes is crucial for risk aware decision making in safety-critical applications. In practice, it remains challenging because trajectory forecasting must jointly model inter-pedestrian interactions to avoid socially/physically implausible futures and capture the inherent multimodality and uncertainty of human motion. To address these challenges, we propose Synergistic Diffusion-adversarial Method (SyDiM), a pioneering generative framework that seamlessly integrates diffusion-based probabilistic modeling with adversarial training strategy in pedestrian trajectory forecasting. Our method leverages diffusion to preserve multimodality and couples it with adversarial social constraints to ensure physically and socially valid interactions. Specifically, we design a Contextual Refinement Module (CRM) within the generator that performs context alignment to capture spatio-temporal dynamics and social context, effectively encoding pedestrian motion patterns while refining group-level interactions in crowded scenes. To enforce social feasibility without sacrificing the multimodality afforded by diffusion, we design an Entity-aware Spatio-temporal Trajectory Discriminator (ESTD). By integrating pedestrian-specific embeddings, it is designed as a feasibility evaluator that uses spatio-temporal attention to identify socially implausible interactions. It then guides the diffusion generator by reweighting its trajectory proposals, effectively suppressing unrealistic trajectories while preserving diversity. To the best of our knowledge, ESTD is the first interaction-conditioned feasibility evaluator tailored for diffusion-based multi-proposal trajectory forecasting. It produces a feasibility score for each predicted trajectory, which is then used to reweight samples during training. Extensive experiments on NBA SportVU, ETH-UCY, and Stanford Drone Datasets demonstrate that our SyDiM achieves consistent performance improvements across all benchmarks. Code will be available at https://github.com/xdclby/SyDiM .

Online action detection and anticipation aim to understand current or upcoming actions in video streams. In industry, current artificial neural network (ANN)-based methods suffer from prohibitive energy consumption, fundamentally limiting their deployment on resource-constrained industrial devices. To bridge the gap between theory and application practice of informatics in industrial environments, we propose a novel knowledge distillation-based spiking neural network (KDSNN), which synergistically integrates bioinspired spike-driven processing with knowledge distillation, significantly reducing the energy consumption. Specifically, KDSNN includes a pioneering spiking neural network (SNN) architecture for online action detection and anticipation, which combines well-designed hierarchical spike convolutional neural network (CNN) block and spike Transformer block to capture spike-driven information. To further improve the performance of our SNN while maintaining low energy consumption, we introduce the knowledge distillation paradigm, which aims to utilize an expert-level ANN as a teacher to guide our SNN. Based on this, we propose a novel distillation loss, which consists of feature distillation and logit distillation. Notably, to address the cross-domain feature alignment in feature distillation, the optimal transport theory is employed to realize cross-domain knowledge transfer for the first time by minimizing the Wasserstein distance between continuous features (ANNs) and discrete features (SNNs). Through extensive evaluations on THUMOS14 and EPIC-Kitchen-100 datasets, the energy consumption of our KDSNN is only 27.1% and 10.0% of the state-of-the-art ANN-based method MAT. Equally importantly, the parameter count of our KDSNN is only 37.0% and 27.7% of MAT on THUMOS14 and EPIC-Kitchen-100, respectively.

The Amodal Instance Segmentation (AIS) task aims to infer the visible and occluded regions of an object instance. Existing AIS methods typically focus on directly predicting visible and occluded regions or leveraging prior knowledge to guide predictions. However, these methods often ignore the perception of occluded views, leading to inaccurate results. To address this issue and achieve high-quality AIS, we propose a boundary-aware Occlusion Perception Network (OPNet). OPNet consists of three main components: the Dynamic Feature Augmentation Pyramid (DFAP), the Dual-path Boundary Aware Module (DBAM), and the Shape-guide Refinement Module (SRM). Specifically, DBAM employs an occlusion-perception strategy to learn discriminative features with boundary information, enabling it to distinguish occlusion from multiple views. Additionally, DFAP and SRM optimize the results by enhancing feature aggregation and imposing geometric constraints. Experiments on the D2SA, KINS, and CWALT datasets show that OPNet significantly outperforms state-of-the-art AIS methods that without prior knowledge. Code is available at https://github.com/ZitengXue/OPNet.

Temporal Action Detection(TAD) is a challenge task in video understanding. The current methods mainly use global features for boundary matching or predefine all possible proposals, while ignoring long context information and local action boundary features, resulting in the decline of detection accuracy. To fill this gap, we propose a Dilation Location Network (DL-Net) model to generate more precise action boundaries by enhancing boundary features of actions and aggregating long contextual information in this paper. Specifically, we design the boundary feature enhancement (BFE) block, which strengthens the actions boundary feature and fuses the similar feature of the different channels by pooling and channel squeezing. Meanwhile, in action location, we design multiple dilated convolutional structures to aggregate long contextual information of time point/interval. We conduct extensive experiments on ActivityNet-1.3 and Thumos14 show that DL-Net is capable of enhancing action boundary features and aggregating long contextual information effectively.