SparseVoxelDet is the first fully sparse object detector for event cameras that never instantiates a dense tensor, achieving 858x GPU memory compression.
March 24, 2026
Original Paper
No Dense Tensors Needed: Fully Sparse Object Detection on Event-Camera Voxel Grids
arXiv · 2603.21638
The Takeaway
By operating exclusively on occupied voxels throughout the backbone and head, it aligns the compute cost with scene dynamics rather than sensor resolution. This is a massive step for deploying high-speed detection on resource-constrained neuromorphic hardware.
From the abstract
Event cameras produce asynchronous, high-dynamic-range streams well suited for detecting small, fast-moving drones, yet most event-based detectors convert the sparse event stream into dense tensors, discarding the representational efficiency of neuromorphic sensing. We propose SparseVoxelDet, to our knowledge the first fully sparse object detector for event cameras, in which backbone feature extraction, feature pyramid fusion, and the detection head all operate exclusively on occupied voxel posi