Are you passionate about #artificial #intelligence, #computer #vision and #health? We are seeking six #PhD candidates for our new AIM lab on AI for Medical Imaging in #Amsterdam: uva.nl/en/content/vac…

ACM #Multimedia 2021: Skeleton-Contrastive 3D Action Representation Learning w/ Fida Mohammad, Hazel Doughty: arxiv.org/abs/2108.03656 We learn invariances to multiple #skeleton representations and introduce various skeleton augmentations via noise contrastive estimation 1/n

#ICCV2021 cam ready: "Social Fabric: Tubelet Compositions for Video Relation Detection" w/ Shuo Chen, Zenglin Shi, Pascal Mettes arxiv.org/abs/2108.08363 TL;DR We classify and detect predicates for pairs of object tubelets a priori, rather than a posteriori. 1/n

I'll be presenting our work #CVPR2025 poster 52b this afternoon. Come chat about recognising adverbs and fine-grained video understanding.

Excited to share our #ECCV2022 work "How Severe is Benchmark-Sensitivity in Video Self-Supervised Learning?" w/ Fida Mohammad, Piyush Bagad and Cees Snoek. We investigate if video self-supervised methods generalize beyond current benchmarks. Details: bpiyush.github.io/SEVERE-website/ 🧵1/5

A PhD student right after they finish finish their PhD

Check out our EXCITING lineup of invited speakers for the #CVPR2025 #CVPR2023 2nd Workshop on Learning with Limited Labelled Data for Image and Video Understanding 🤗 sites.google.com/view/l3d-ivu-2…

As soon as we hear about the decision release from our #ICCV2023 Program Chairs we will let you know! In the meantime …

The list of paper IDs for accepted #ICCV2023 papers is now available at the following link: drive.google.com/file/d/1t0XHej…

Enhanced version of this paper accepted to #ICCV2023. 🥳🥳 ps: my belated birthday gift!

Happy our paper 'Tubelet-Contrastive Self-Supervision for Video-Efficient Generalization' was accepted to #ICCV2023. Congrats Fida Mohammad! Preprint: arxiv.org/abs/2303.11003

How to do research: people.csail.mit.edu/billf/publicat… Summary : 1. Slow down to speed up. 2. Find why something does not work as much as possible. 3. Hard working is necessary . 4. Collaborate. Small piece of a big cake> Big piece of small cake. 5. Communicate effectively. #icvss23

Next Thursday at #ICCV2023, we'll present our work on self-supervised learning for motion-focused video representations. Work w/ Fida Mohammad and Cees Snoek We learn similarities between videos with identical local motion dynamics but an otherwise different appearance. 1/6

📢SIGMA: Sinkhorn-Guided Masked Video Modeling got accepted to European Conference on Computer Vision #ECCV2026 #ECCV2024 TL;DR: Instead of using pixel targets in Video Masked Modeling, we reconstruct jointly trained features using Sinkhorn guidance, achieving SOTA. 📝Project page: quva-lab.github.io/SIGMA/ 🌐Paper:

Checkout our poster if you are at #ECCV2024

Our #ACCV2024 Oral "LocoMotion: Learning Motion-Focused Video-Language Representations" w/ Fida Mohammad and Cees Snoek is now on ArXiv arxiv.org/abs/2410.12018 We remove the spatial focus of video-language representations and instead train representations to have a motion focus.

Super excited to share that 4 papers from our KAUST Computer Vision Lab (IVUL) IVUL lab got into #CVPR2025! Huge congrats to all the authors and our amazing collaborators! 🎉👏 #CVPR2025 KAUST Research

Just released "DiffCLIP", extending Differential Attention proposed by Tianzhu Ye to CLIP models - replacing both visual & text encoder attention with the differential attention mechanism! TL;DR: Consistent improvements across all tasks with only 0.003% extra parameters!

We have had 4 deaths, not 6/7 as some channels/websites are insisting & all injured are stable 'n out of danger. Even 4 is 4 too many 2/3