Search papers, labs, and topics across Lattice.
3
0
2
Robot skills degrade significantly (56% success drop) from minor starting state perturbations, yet isolated skill evaluation reveals substantial per-skill performance gaps hidden by aggregate metrics.
Achieving high-quality video object scaling without the overhead of 3D reconstruction, ScaleVid redefines the boundaries of real-time video synthesis.
By explicitly aligning intra-modality relationships, SSR$^2$-GCD unlocks more effective cross-modal representation learning for generalized category discovery.
