Congratulations to our ESR 5 from MENELAOSแดบแต, Saqib Nazir, who successfully defended his PhD thesis on the topic of “Deep Depth from Defocus (Deep DFD) for near range and in-situ 3D exploration” on 4 December, 2023 at the University POLITEHNICA of Bucharest under the supervision of Prof. Dr. Ing. Daniela COLTUC.
An image taken with a conventional camera is a 2D projection of a 3D scene. During the imaging process, the information of the 3rd dimension i.e., the scene depth is lost. However, it can be recovered by computation, from the set of visual cues present in the images. This has created the premises to perceive the world in 3D by means of conventional cameras.
The thesis addresses the challenge of monocular depth estimation using a single defocused image, a pivotal task in computer vision given the wide area of applications ranging from robotics to self-driving cars. The recent advances in deep learning has
revolutionized the field of computer vision and in particular, depth estimation. However, prior deep learning-based techniques often neglect the potential of defocus blur, which is an important cue for depth estimation. The existing solutions employ multiple images or focal stacks of the same scene and rarely single images. To fill in this gap, we propose a novel architecture called 2HDED:NET, that addresses both depth estimation and image deblurring from a single defocused image.
Due to the absence of datasets containing naturally defocused images and depth ground truth, networks like 2HDED:NET are typically trained on synthetic data, a fact that reduces their performances on real data. In order to train 2HDED:NET to its full potential, we proposed a new dataset called iDFD containing naturally defocused images, double annotated with the all-in-focus image, and the Time of Flight depth map.
Finally, the thesis explores the promising field of self-supervised learning, by con- verting 2HDED:NET to defocus map estimation in the absence of ground truth depth for training. To accomplish this, 2HDED:NET is enhanced with a defocus simulation module that reconstructs the defocused image from the all-in-focus one and the estimated defocus map. The modelโs proficiency in defocus map estimation is on par with that of state-of-the-art supervised models that use multiple images. Comprehensive experiments conducted on various real or synthetic datasets validated the efficacy of the proposed
approaches for depth or defocus map estimation in various settings, encompassing indoor and outdoor environments.
Recent Comments