Elisa Ricci

Data mixing, especially with instruction-heavy data, emerges as the crucial factor for optimizing VLM training, challenging traditional filtering approaches.

Skip the labeled data: TEGU uses LLM-derived textual information to achieve state-of-the-art zero-shot temporal action localization.

Predefined interaction vocabularies are holding back HOI detection, but MLLMs can unlock truly unconstrained understanding of how humans and objects interact.

Unlock human-interpretable video understanding without task-specific training: TF-SMOT leverages off-the-shelf vision-language models to achieve state-of-the-art semantic multi-object tracking.

Unlocking fairer vision-language models may be as simple as intervening in the sparse latent space of a sparse autoencoder, enabling targeted bias removal without harming performance.

Forget specialized classifiers—LMMs, enhanced with in-context learning and a novel iterative refinement method (CIRCLE), can outperform even fine-tuned VLMs in both closed and open-world classification.