Event-Based Vision Advances

The field of event-based vision is rapidly advancing, with a focus on improving image enhancement, super-resolution, and object recognition. Researchers are exploring new methods to fully exploit the advantages of event cameras, such as their high dynamic range and low latency. One notable direction is the development of novel neural network architectures that can effectively process event data, including spiking neural networks and graph neural networks. These advancements have significant implications for various applications, including surveillance, augmented reality, and robotics. Noteworthy papers include:

  • A paper proposing a transfer learning framework for event-based facial expression recognition, which achieves a 93.6% recognition rate on the e-CK+ database.
  • A paper introducing a self-supervised pre-training framework for noisy and sparse events, which consistently outperforms state-of-the-art methods on various downstream tasks.

Sources

Exploring Fourier Prior and Event Collaboration for Low-Light Image Enhancement

Ultralight Polarity-Split Neuromorphic SNN for Event-Stream Super-Resolution

evTransFER: A Transfer Learning Framework for Event-based Facial Expression Recognition

STARE: Predicting Decision Making Based on Spatio-Temporal Eye Movements

Drone Detection with Event Cameras

A deep learning approach to track eye movements based on events

Will You Be Aware? Eye Tracking-Based Modeling of Situational Awareness in Augmented Reality

Revealing Latent Information: A Physics-inspired Self-supervised Pre-training Framework for Noisy and Sparse Events

Built with on top of