Bachelor and Master Theses

In computer vision unsupervised object segmentation aims to identify distinct objects or parts of objects in an image purely based on visual cues, e.g. color, texture, depth, structural consistency, and motion in videos. State-of-the art methods are based on self-supervised or representation learning. However, most of these unsupervised methods segment object-like regions, but they lack true semantic understanding.

The aim of thesis is to develop a self-supervised framework for unsupervised object segmentation ensuring temporal consistency and stability while maintaining semantic fidelity.

Expected outcome

-              A novel unsupervised object segmentation framework.

-              Improved temporal stability in video sequences.

-              Implementation with benchmark evaluation.

-              Temporal feature learning using unlabelled image/video dataset.

• Python programming and PyTorch
• Knowledge on image processing, CNNs and CNN architectures  
• Working with benchmark image datasets, e.g COCO, CIFAR, ImageNet