The field of computer vision and multimodal learning is moving towards more robust and efficient representations of complex data. Recent research has focused on leveraging auxiliary information, such as visual attributes and temporal context, to improve retrieval performance and bridge the semantic gap between different modalities. Another significant direction is the development of event-driven vision methods, which can efficiently process and represent asynchronous event streams from event cameras. These methods have shown great promise in various applications, including person re-identification, object recognition, and visible-infrared person re-identification. Notable papers in this area include S3CE-Net, which proposes a spike-guided spatiotemporal semantic coupling and expansion network for long-sequence event-based person re-identification, and BiMa, which introduces a novel framework to mitigate biases in text-video retrieval via scene element guidance. Additionally, the proposed dataset FRED provides a valuable resource for researchers to explore drone detection, tracking, and trajectory forecasting using event cameras.
Advances in Multimodal Representation and Event-Driven Vision
Sources
Leveraging Auxiliary Information in Text-to-Video Retrieval: A Review
S3CE-Net: Spike-guided Spatiotemporal Semantic Coupling and Expansion Network for Long Sequence Event Re-Identification
SA-Person: Text-Based Person Retrieval with Scene-aware Re-ranking
Zero-Shot Chinese Character Recognition with Hierarchical Multi-Granularity Image-Text Aligning
Video-Level Language-Driven Video-Based Visible-Infrared Person Re-Identification
Probabilistic Online Event Downsampling
BiMa: Towards Biases Mitigation for Text-Video Retrieval via Scene Element Guidance
ROSA: Addressing text understanding challenges in photographs via ROtated SAmpling
CoLa: Chinese Character Decomposition with Compositional Latent Components
Learning from Noise: Enhancing DNNs for Event-Based Vision through Controlled Noise Injection
EV-Flying: an Event-based Dataset for In-The-Wild Recognition of Flying Objects
Person Re-Identification System at Semantic Level based on Pedestrian Attributes Ontology
Spike-TBR: a Noise Resilient Neuromorphic Event Representation
Beyond Cropped Regions: New Benchmark and Corresponding Baseline for Chinese Scene Text Retrieval in Diverse Layouts
FRED: The Florence RGB-Event Drone Dataset