Person

David Mizrahi

About this result
This page is automatically generated and may contain information that is not correct, complete, up-to-date, or relevant to your search query. The same applies to every other page on this website. Please make sure to verify the information with EPFL's official sources.
Related publications (3)

Please note that this is not a complete list of this person’s publications. It includes only semantically relevant works. For a full list, please refer to Infoscience.

Composite Relationship Fields with Transformers for Scene Graph Generation

Alexandre Massoud Alahi, David Mizrahi, George Adaimi

Scene graph generation (SGG) methods extract relationships between objects. While most methods focus on improving top-down approaches, which build a scene graph based on detected objects from an off-the-shelf object detector, there is a limited amount of w ...
2023

4M: Massively Multimodal Masked Modeling

Shuqing Teresa Yeo, Amir Roshan Zamir, Oguzhan Fatih Kar, Roman Christian Bachmann, David Mizrahi

Current machine learning models for vision are often highly specialized and limited to a single modality and task. In contrast, recent large language models exhibit a wide range of capabilities, hinting at a possibility for similarly versatile models in co ...
Neural Information Processing Systems (Nips)2023

MultiMAE: Multi-modal Multi-task Masked Autoencoders

Amir Roshan Zamir, Roman Christian Bachmann, David Mizrahi, Andrei Atanov

We propose a pre-training strategy called Multi-modal Multi-task Masked Autoencoders (MultiMAE). It differs from standard Masked Autoencoding in two key aspects: I) it can optionally accept additional modalities of information in the input besides the RGB ...
SPRINGER INTERNATIONAL PUBLISHING AG2022