Mina Ghadimi Atigh

I am a Ph.D. student at the University of Amsterdam. I'm working on "Hyperbolic visual representation learning" under the supervision of Pascal Mettes at VisLab. I recently finished my internship at Apple MLR team, working with Xavier Suau.

I completed my Master's degree in the Statistical Machine Learning Lab at the Department of Computer Engineering and Information Technology, Amirkabir University of Technology, Tehran, Iran, in 2019. I received my B.Sc. degree from the Department of Computer Engineering and Information Technology, Amirkabir University of Technology, Tehran, Iran, in 2016.

Email  /  CV  /  Google Scholar  /  Twitter  /  Github  /  LinkedIn

I'm interested in computer vision and understanding images and videos. I'm interested in exploring images and video understanding in non-Euclidean spaces, such as hyperbolic space.

In this survey, We outline how hyperbolic learning is performed in all themes and discuss the main research problems that benefit from current advances in hyperbolic learning for computer vision.

Hyperbolic Image Segmentation opens up new possibilities and practical benefits for segmentation, such as uncertainty estimation and boundary information for free, zero-label generalization, and increased performance in low-dimensional output embeddings.

Where existing works on prototype-based learning in hyperbolic space requires prior knowledge to operate, Hyperbolic Busemann Learning places prototypes at the ideal boundary of the Poincar`e ball. This enables a prototype positioning without the need for prior knowledge.

We propose a Convolutional Relational Machine for group activity recognition by extracting the relationships between persons. We show that the activity map is a useful representation that effectively encodes the spatial relations.

In this paper, we presented multi-stream convolutional networks as a framework for group activity recognition. In this framework, new modalities both in spatial and temporal domains can be easily plugged in to improve the model’s power.

In this paper, we proposed a novel method for combining cues of different body regions for the task of person recognition in photo album. Our approach uses two distinct recurrent neural networks to extract information present in different parts of a human photo in order to improve recognition performance.

Thanks Jon for the template!