Description:

Transparent objects are very common in our life but always ignored in computer vision research. Multimodality inputs such as polarization images help to recognize transparent objects [1] [2]. In the scope of this thesis, we will extend our self-supervision pipeline [3] to transparent categories leveraging polarization images.

Requirement:

Students should be familiar with pytorch or tensorflow and have hands-on experience with deep learning in past projects. Aiming for a paper publication would be a plus.


width:20px

Glass Segmentation [1]

LatentFusion[2]

Polarmetric Pose Prediction [2]

Benefits:

you can learn state of art 6d object pose estimation networks and cooperate with experts both at the Chair and inside our industrial corporate partners.

[1] Kalra, Agastya, et al. “Deep polarization cues for transparent object segmentation.” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020.

[2] Gao, Daoyi, et al. “Polarimetric Pose Prediction.” arXiv preprint arXiv:2112.03810 (2021).

[3] Manhardt, Fabian, et al. “CPS++: Improving class-level 6D pose and shape estimation from monocular images with self-supervised learning.” arXiv preprint arXiv:2003.05848 (2020).