Homepage of Pascal Mettes, University of Amsterdam

About

My research strives to find a bridge between prior knowledge and deep networks. Deep learning in computer vision thrives under examples but commonly ignores additional explicit knowledge about the research problem. Think about hierarchical relations between categories, relational knowledge between tasks, or spatio-temporal knowledge. The research of me and my team focuses on discovering the shared geometry between pixels and knowledge. Specifically, we research hyperbolic and hyperspherical geometry for deep learning and learning with prototypes. Advances in these research directions are investigated on various computer vision problems, including but not limited to hierarchical recognition, search, localisation, and zero-shot recognition in images and videos.

News

2022

[Talk] Talk at Utrecht University. Host: Ronald Poppe (May).
[Talk] Talk at TU Delft. Host: Jan van Gemert (May).
[Accepted paper] CVIU: "Are 3D Convolutional Networks Inherently Biased Towards Appearance?" [pdf].
[Accepted paper] ICMR 2022: "Teaching a New Dog Old Tricks: Contrastive Random Walks in Videos with Unsupervised Priors" [pdf coming soon].
[Accepted paper] CVPR 2022: "Hyperbolic Image Segmentation" [pdf].

2021

[Talk] Talk at Efficient Deep Learning Winterschool on "Hyperbolic Deep Learning". Oestgeest, the Netherlands.
[Accepted paper] BMVC 2021: "Zero-Shot Action Recognition from Diverse Object-Scene Compositions" [pdf].
[Accepted paper] BMVC 2021: "Diagnosing Errors in Video Relation Detectors" [pdf].
[Accepted paper] NeurIPS 2021: "Hyperbolic Busemann Learning with Ideal Prototypes" [pdf].
[Accepted paper] ICCV 2021: "Social Fabric: Tubelet Compositions for Video Relation Detection" [pdf].
[Accepted paper] TIP: "Unsharp Mask Guided Filtering" [pdf].
[Accepted paper] MICCAIw 2021: "Frequency-Supervised MR-to-CT Image Synthesis" [pdf].
[Talk] Talk at TU Dresden. Host: prof. Martin Keller-Ressel, Dresden Germany (June).
[Talk] Talk at Explainable Machine Learning group. Host: prof. Zeynep Akata, Tübingen Germany (June).
[Organization] ICCV 2021: Workshop on Structured Representations for Video Understanding [link].
[Vacancy] PhD position on action and intention understanding in videos [closed].
[Accepted paper] CVPR 2021: "Few-Shot Transformation of Common Actions into Time and Space" [pdf].
[Accepted paper] IJCV: "Object Priors for Classifying and Localizing Unseen Actions" [pdf].
[Talk] Talk at Diagnostic Image Analysis Group, Radboudumc (January).

2020

[Accepted paper] 3DV 2020: "Adversarial Self-Supervised Scene Flow Estimation" [pdf].
[Media] Coverage of our research in the New Scientist [link].
[Accepted paper] ECCV 2020: "Localizing the Common Action Among a Few Videos" [pdf].
[Accepted paper] ECCV 2020: "PointMixup: Augmentation for Point Clouds" [pdf].
[Talk] Speaker at Bertelsmann Video AI event (July).
[Accepted paper] CVIU: "Open cross-domain visual search" [pdf].
[Accepted paper] CVPR 2020: "Searching for Actions on the Hyperbole" [pdf].
[Organization] Area chair for BMVC 2020.
[Accepted paper] ICMR 2020 oral: "Heterogeneous Non-Local Fusion for Multimodal Activity Recognition" [pdf].
[Accepted paper] ICMR 2020 oral: "Interactivity Proposals for Surveillance Videos" [pdf].
[Accepted paper] ICMR 2020 oral: "Trajectory Prediction Network for Future Anticipation of Ships" [pdf].
[Talk] Speaker at VISUM summer school 2020 (July).
[Accepted paper] ACM TOMM: "Shuffled ImageNet-Banks for Video Event Detection and Search" [pdf].
[Vacancy] PhD position on knowledge graph of a video [closed].
[Talk] Public talk at Aalto University in Finland (January).
[Vacancy] PhD position on action understanding [closed].

2019

[Talk] Public talk at the NVPHBV fall meeting (November).
[Accepted paper] NeurIPS 2019: "Hyperspherical Prototype Networks" [pdf].
[Accepted paper] ICCV 2019: "Counting with Focus for Free" [pdf].
[Accepted paper] ICCV 2019: "SILCO: Show a Few Images, Localize the Common Object" [pdf].
[Accepted paper] ICCV workshops 2019: "4-Connected Shift Residual Networks" [pdf].
[Accepted paper] ICCV workshops 2019: "Bayesian 3D ConvNets for Action Recognition from Few Examples" [pdf].
[Talk] Speaker at VISUM summer school 2019 (July).