A modified automatic label generation method under low-light environments for event-based semantic segmentation

Event cameras are novel bio-inspired sensors, which have the advantages of high temporal resolution, high dynamic range and low latency compared with traditional cameras. Employing event cameras for semantic segmentation is an area that deserves in-depth research to exploit the great potential of event cameras. Due to the asynchronous nature of data from event cameras, it is challenging to produce accurate labels directly from events for the training of event-based semantic segmentation networks, especially in low-light environments. In this paper, we propose a new automatic label generation method under low-light environment for event-based semantic segmentation. Firstly we propose a modified automatic label generation module to generate more accurate labels for training. Meanwhile, we propose an event-assisted low-light enhancement module to improve the accuracy of the labels of low-light data. Finally we provide an event-based semantic segmentation dataset with high-precision paired labels for both low-light and normal environments. Experiments have shown that our method accomplishes high-quality event-based semantic segmentation in both low-light and normal environments by acquiring high-precision labels. To the best of our knowledge, our work is the first to realize event-based semantic segmentation in low-light environments in the field of event-based semantic segmentation.