Search papers, labs, and topics across Lattice.
3
0
4
0
Freezing SAM3 and simply concatenating images spatially achieves state-of-the-art few-shot segmentation, revealing that strong cross-image reasoning can emerge from surprisingly simple formulations.
Dramatically improve HDR reconstruction from single LDR images, even in overexposed regions, without any training by cleverly combining diffusion models and existing reconstruction pipelines.
By predicting tracking models rather than image features, GOT-JEPA unlocks more robust object tracking, even when objects are heavily occluded or the environment is dynamic.
