The field of video understanding and analysis is rapidly evolving, with a focus on developing innovative methods for video anomaly detection, event prediction, and temporal grounding. Recent research has explored the use of large language models, graph-based transformers, and reinforcement learning to improve the accuracy and efficiency of video analysis tasks. Notably, the integration of multimodal learning and spatio-temporal reasoning has shown promising results in capturing complex temporal dynamics and relationships between video frames. Furthermore, the development of new datasets and benchmarks has facilitated the evaluation and comparison of different models, driving progress in the field. Some noteworthy papers include MoniTor, which introduces a novel online video anomaly detection method using large language models, and EventFormer, which proposes a graph-based transformer approach for action-centric video event prediction. Overall, the field is moving towards more sophisticated and effective methods for video understanding and analysis, with potential applications in various areas such as surveillance, healthcare, and education.
Advancements in Video Understanding and Analysis
Sources
MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection
EventFormer: A Node-graph Hierarchical Attention Transformer for Action-centric Video Event Prediction
Frame-Difference Guided Dynamic Region Perception for CLIP Adaptation in Text-Video Retrieval
Human-Centric Anomaly Detection in Surveillance Videos Using YOLO-World and Spatio-Temporal Deep Learning
TrajGATFormer: A Graph-Based Transformer Approach for Worker and Obstacle Trajectory Prediction in Off-site Construction Environments
Accident Anticipation via Temporal Occurrence Prediction
VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree
HieraMamba: Video Temporal Grounding via Hierarchical Anchor-Mamba Pooling
Evaluation of Vision-LLMs in Surveillance Video
VideoTG-R1: Boosting Video Temporal Grounding via Curriculum Reinforcement Learning on Reflected Boundary Annotations
Video-Thinker: Sparking "Thinking with Videos" via Reinforcement Learning
EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT
YTLive: A Dataset of Real-World YouTube Live Streaming Sessions
StreamingCoT: A Dataset for Temporal Dynamics and Multimodal Chain-of-Thought Reasoning in Streaming VideoQA
Enhancing Temporal Understanding in Video-LLMs through Stacked Temporal Attention in Vision Encoders
EgoExo-Con: Exploring View-Invariant Video Temporal Understanding
Which Way Does Time Flow? A Psychophysics-Grounded Evaluation for Vision-Language Models
LoCoT2V-Bench: A Benchmark for Long-Form and Complex Text-to-Video Generation
Are Video Models Ready as Zero-Shot Reasoners? An Empirical Study with the MME-CoF Benchmark