Search papers, labs, and topics across Lattice.
Affiliation:
4
0
6
10
State-of-the-art video object removal methods achieve high visual fidelity but systematically fail to maintain causal consistency in real scenes.
Hyperspherical latent spaces unlock better 3D scene understanding from vision transformers, especially when bandwidth is constrained.
Forget scaling laws: teaching models *how* to think like a cartoonist unlocks expert-level humor understanding, even surpassing larger black-box models.
Instruction-tuned MLLMs, despite excelling in zero-shot scenarios, surprisingly fail to leverage few-shot examples or chain-of-thought prompting, suggesting a fundamental limitation in their ability to learn from demonstrations.
