Data augmentation for multi-view people tracking

MOT20 tracking challenge
MOT20 tracking challenge
Ground plane projection
Ground plane projection

When training deep learning model data augmentation is commonly used to improve generalization. There exist a wide type of data augmentation, the most common ones are scaling, cropping, flipping and rotation.

While it is straightforward to apply data augmentation in a single view classification setting, data augmentation in multi-view setup can be challenging. It needs to take into account camera calibration and update it accordingly such that alignment between view is preserved.

The goal of this project is to develop and implement different types of multi-view data augmentation. The proposed augmentation will then be evaluated on the task of multi-view multi-object tracking.

Prerequisites

The candidate should have Python programming experience. Previous experience with deep learning and PyTorch, is a plus. Knowledge about image transformation and camera model is recommended.

Contact

For further information contact: