Search papers, labs, and topics across Lattice.
8
0
8
FLEET achieves superior performance in reinforcement learning with event cameras by directly processing event sequences, challenging the limitations of traditional CNN-based approaches.
VDE can pinpoint the optimal visual layer for VLMs with remarkable accuracy, saving researchers from costly exhaustive searches.
MATCH achieves state-of-the-art anomaly detection and segmentation performance while running efficiently on consumer-level hardware, revolutionizing multi-view analysis in industrial applications.
RL slashes ion shuttling operations by over 36% in trapped-ion quantum computers, outperforming current heuristics.
Unlock human-interpretable image classifications from frozen DINOv2 models with DINO-QPM, which surprisingly surpasses DINOv2 linear probe accuracy while providing superior explanation quality.
Achieve 2x the efficiency of standard patch pruning by intelligently dropping video tokens in early transformer layers, guided by temporal context.
Normalizing flows can flag anomalous relationships in scene graphs with 10% better accuracy and 5x faster speed than existing methods, while also exhibiting superior robustness to semantic variations.
Fine-grained foot motion capture, a notoriously hard problem, gets a 30% accuracy boost by cleverly lifting 2D keypoints to 3D using motion capture data and contextual information, bypassing the need for direct image-3D annotation pairs.
