Search papers, labs, and topics across Lattice.
2
0
5
Generalizing to unseen compositions? This plug-and-play method leverages structure in the embedding space to adapt prompts, significantly boosting open-vocabulary zero-shot learning.
By learning to intelligently "zoom in" on relevant image regions, TikArt significantly boosts MLLM performance on fine-grained visual reasoning tasks.
