Oriane Simeone from Inria gave a fantastic talk today. Combing lightweight local features with activation maps to achieve dramatic gains for image retrieval!

18-month Post-doc position at CEA List (Paris, France) and Inria (Rennes, France): Multimodal entity representation and disambiguation meerqat.fr/wp-content/upl…

Attending #ICLR2022 this week? Come chat with Shashank, me and Yannis Avrithis tomorrow Thursday 28/4 at our poster - "It Takes Two to Tango: Mixup for Deep Metric Learning". We will be at the virtual spot - G2 (poster session 10) between 11:30 to 13:30 CET. iclr.cc/virtual/2022/p…

Do CNNs and ViTs really need to have different forms of pooling? We discard the CLS in ViTs and introduce "SimPool" at the end of the network. Whether in supervised or self-supervised setting, "SimPool" produces high-quality attention maps and has great localization properties.

I will be presenting MultiMix at #NeurIPS2023 on Tuesday 12th Dec during Poster session 1 (10:45 - 12:45 CDT). Will be at Poster #1001, come say Hi :) 👋 More details can be found from the poster below or from our paper - arxiv.org/pdf/2311.05538… Joint work with Yannis Avrithis

Really happy to share that DoRA is accepted as an Oral to ICLR 2026 #ICLR2024 Using just “1 video” from our new egocentric dataset - Walking Tours, we develop a new method that outperforms DINO pretrained on ImageNet on image and video downstream tasks. More details in 🧵👇

Congrats Shashank, Mamshad Nayeem Rizve, joao carreira and Yuki!