Description
30 credits - 3D scene reconstruction from images by deep neural networks
Background
Scania is now undergoing a transformation from being a supplier of trucks, buses and engines to a supplier of complete and sustainable transport solutions.
This thesis work will lie under the supervision of the research group EARA, which develops the deep learning-based methods that are used in the scene perception for autonomous driving.
Description
Deep neural networks are outperforming many classical approaches for computer vision tasks such as semantic segmentation and object detection. The inverse graphic problem - consisting in retrieving the 3D geometry of a scene and its appearance from given images and camera poses - is also now addressed by deep learning. The recent emergence of NeRF (Neural Radiance Fields) techniques showed promising results in this 3D reconstruction task. Such technology could help us for many use cases such as propagating annotation from one image to other images (auto-labelling) or generating new 2D/3D environments for testing purposes.
This master’s thesis will focus on investigating the ability of Neural Fields to generate realistic 3D reconstruction from images in an offline manner and this work will most certainly entail the following steps:
- Perform a literature review on the Neural Radiance Fields for dynamic scene reconstruction. A particular focus is expected on how dependent such techniques on calibration of the input images are.
- Experiment with the selected approach on our in-house dataset and identify the relevant metrics to assess the performance of the method
- Most of the papers propose methods that require per-scene optimization on limited spatial environment. Based on papers in the literature that address these limitations, you propose a way to overcome those by considering the specifications of the autonomous vehicle domain
Applicants
One thesis worker studying a master's program in Computer Science, Electrical Engineering or similar. Applicants are expected to have a good understanding of computer vision, machine learning and practice thereof. The applicant should have sufficient software development knowledge to be able to implement/analyse mathematical concepts. Prior experience with deep learning or computer graphics is a plus. The applicant should be able to work in a diverse environment and communicate effectively in English. The personal traits of being agile, giving/receiving constructive feedback and taking initiatives will come in handy.
Time plan
The project is planned for 20 weeks and can be started any time in early Spring 2023.
Applicants will be assessed on a continuous basis until the position is filled.
**Contacts