Detecting Every Object in Events: A Robust and Efficient Approach for Class-Agnostic Open-World Object Detection

The proposed Detecting Every Object in Events (DEOE) approach leverages the unique characteristics of event cameras, such as sub-millisecond latency and high dynamic range, to achieve robust and efficient class-agnostic open-world object detection. DEOE utilizes spatio-temporal consistency and task disentanglement to identify and incorporate potential unknown objects during training, enhancing the model's generalization capabilities.