1

Conjuring Positive Pairs for Efficient Unification of Representation Learning and Image Synthesis
Imanol G. Estepa, Jesús M. Rodríguez-de-Vera, Ignacio Sarasúa, Bhalaji Nagarajan, Petia Radeva
Cite ArXiv
Conjuring Positive Pairs for Efficient Unification of Representation Learning and Image Synthesis
CLIP-DoRA: Weight-decomposed Low-rank Adaptation for Efficient Vision-Language Models
Foundational Vision-Language (V-L) models, such as CLIP, have progressed computer vision research by providing general “shared” latent …
Jesús M. Rodríguez-de-Vera, Imanol G. Estepa, Bhalaji Nagarajan, Petia Radeva
Cite ScienceDirect
LOFI: LOng-tailed FIne-Grained Network for Food Recognition
Food recognition plays a crucial role in several healthcare applications. Nevertheless it presents significant computer vision …
Jesús M. Rodríguez-de-Vera, Imanol G. Estepa, Marc Bolaños, Bhalaji Nagarajan, Petia Radeva
Cite CVF Open Access
Dining on Details: LLM-Guided Expert Networks for Fine-Grained Food Recognition
Dining on Details (DoD) is an innovative fine-grained food classification approach using large language models to sort dataset classes into subsets. Powered by the robust ImageBind embedding space, DoD excels in distinguishing similar classes. Universally compatible, DoD integrates seamlessly with any existing classification architecture. Extensive testing on various food datasets and backbones shows performance boosts of 0.5% to 1.61%, and even achieves SoTA results on the Food-101 dataset.
PWC
Jesús M. Rodríguez-de-Vera, Pablo Villacorta, Imanol G. Estepa, Marc Bolaños, Ignacio Sarasúa, Bhalaji Nagarajan, Petia Radeva
Cite DOI
Dining on Details: LLM-Guided Expert Networks for Fine-Grained Food Recognition
Good Fences Make Good Neighbours
Our new framework, “Mending Neighbours,” improves contrastive learning by effectively identifying and replacing unhelpful ’neighbour’ representations with our unique “Bridge Points”. This unsupervised method enhances model performance by ensuring only the most informative representations are learned. Outperforming leading methods on three benchmark datasets, our work underscores the crucial role of careful neighbour selection in contrastive learning.
Imanol G Estepa, Jesús M. Rodríguez-de-Vera, Bhalaji Nagarajan, Petia Radeva
Cite OpenReview
Good Fences Make Good Neighbours