Search papers, labs, and topics across Lattice.
3
0
3
Masking critical query tokens in diffusion transformers can dramatically enhance the detail and quality of generated images.
Frequency-aware embeddings reveal that separating style from content can dramatically enhance image customization fidelity and text alignment.
By explicitly modeling modality reliability, RSGMamba avoids feature degradation from noisy or misaligned data, achieving state-of-the-art multimodal semantic segmentation.
