Search papers, labs, and topics across Lattice.
Shanghai Jiao Tong University
7
0
9
Directly conditioning video diffusion on audio wastes massive capacity on static background and identity pixels鈥攔outing control transitively through a causal motion latent distilled under a single frozen video teacher achieves real-time streaming at 15.4 FPS with zero fidelity loss.
Achieving state-of-the-art performance in real-world image super-resolution while maintaining a rapid single-step processing time and minimal computational overhead.
Current AI systems struggle to conduct independent scientific research, with performance plummeting by nearly 50% when human guidance is removed.
Prioritizing video token importance can dramatically enhance reconstruction quality, even in the face of severe packet loss.
Achieving superior rate-distortion performance, GenSplatCodec redefines low-bitrate Gaussian compression by seamlessly integrating geometry and appearance cues for high-fidelity 3D scene reconstruction.
Transforming video generation from a pixel sampling problem to a structured orchestration of the physical world, WNM enables unprecedented control and efficiency in content creation.
ClinHallu reveals that pinpointing the exact source of hallucinations in medical MLLM reasoning can drastically enhance model reliability.
