Future Interaction @ Samsung AI Center Cambridge

[CVPR'26] Restore, Assess, Repeat: A Unified Framework for Iterative Image Restoration

[CVPR'26] No Hard Negatives Required: Concept Centric Learning Leads to Compositionality without Degrading Zero-shot Capabilities of Contrastive Models

Multi-scale Image Super Resolution with a Single Auto-Regressive Model

[ICLR'23] Effective Self-supervised Pre-training on Low-compute networks without Distillation

[EMNLP Findings 2024] MobileQuant: Mobile-friendly Quantization for On-device Language Models

[ICCV 2023] Black Box Few-Shot Adaptation for Vision-Language models

[ICCV 2023] Bayesian Prompt Learning for Image-Language Model Generalization

[ECCV 2022] EdgeViT: Competing Light-weight CNNs on Mobile Devices with Vision Transformers

[NeurIPS 2021] Low-Fidelity Video Encoder Optimization for Temporal Action Localization

[NeurIPS 2021] Space-time Mixing Attention for Video Transformer