Generative Event Pretraining with Foundation Model Alignment
This work addresses the problem of robust visual feature learning for event cameras, which is crucial for applications in fast motion and challenging illumination, but it is incremental as it builds on existing pretraining and alignment techniques.
The paper tackled the challenge of training event-based visual foundation models due to unique sensing characteristics and limited labeled data by proposing GEP, a two-stage framework that transfers semantic knowledge from image datasets to event data and learns event-specific temporal dynamics, resulting in outperforming state-of-the-art methods on tasks like object recognition, segmentation, and depth estimation.
Event cameras provide robust visual signals under fast motion and challenging illumination conditions thanks to their microsecond latency and high dynamic range. However, their unique sensing characteristics and limited labeled data make it challenging to train event-based visual foundation models (VFMs), which are crucial for learning visual features transferable across tasks. To tackle this problem, we propose GEP (Generative Event Pretraining), a two-stage framework that transfers semantic knowledge learned from internet-scale image datasets to event data while learning event-specific temporal dynamics. First, an event encoder is aligned to a frozen VFM through a joint regression-contrastive objective, grounding event features in image semantics. Second, a transformer backbone is autoregressively pretrained on mixed event-image sequences to capture the temporal structure unique to events. Our approach outperforms state-of-the-art event pretraining methods on a diverse range of downstream tasks, including object recognition, segmentation, and depth estimation. Together, VFM-guided alignment and generative sequence modeling yield a semantically rich, temporally aware event model that generalizes robustly across domains.