Presents a causal spatiotemporal convolutional network for event-camera eye tracking on resource-constrained edge hardware, using architectural simplification, inference buffering, and affine augmentation to compensate for scarce training data. The model placed at the top of the AIS 2024 event-based eye tracking challenge with 0.9916 p10 accuracy on the Kaggle private test set while sustaining over 90% activation sparsity.

Full access to tools, docs, models, and all future white papers. Free to join. No spam. Unsubscribe anytime.