Search papers, labs, and topics across Lattice.
Harbin Institute of Technology, Shenzhen Loop Area Institute
6
0
6
Mainstream models falter in multi-reference image generation, but DyRef's innovative training framework boosts their performance significantly.
ChartArena reveals that even top proprietary models struggle with diagrammatic structures, exposing critical gaps in current chart parsing capabilities.
Reasoning across languages doesn't have to break the bank: a new framework slashes token costs by over 50% while maintaining accuracy, especially boosting performance in low-resource languages.
Video-ToC drastically improves video understanding by forcing Video LLMs to focus on relevant visual cues, leading to state-of-the-art performance and reduced hallucinations.
Amodal SAM unlocks zero-shot amodal segmentation by adapting the Segment Anything Model (SAM) to infer complete object shapes, even in the presence of occlusions.
Stop treating human motion as a series of discrete actions: TranCLR learns smoother, more accurate skeleton representations by explicitly modeling the continuous transitions between actions.
