Semi-supervised iterative teacher-student learning for monocular depth estimation

Süvari, Cemal Barışkan
Advances in robotics area and autonomous vehicles have increased the need for accurate depth measurements. Depth estimation is one of the oldest problems of computer vision area. While the depth can be estimated by using many methods, finding a cheap and efficient way of doing it was studied for many years. Although, depth measurements using Lidar sensors or RGB-D cameras provides accurate results, due to cost and narrow applicability they are not very effective. On the other hand, using deep learning architectures to estimate depth seems to provide a more efficient, cheaper and robust solution compared to other methods. With the progress in deep learning, monocular depth estimation problem has gained a lot of attention. Recently, representation learning methods showed very promising accuracy results in depth estimation from single images. In this thesis, a deep learning based network architecture is proposed for monocular depth estimation problem. Furthermore, the network is trained with an iterative teacher-student learning framework in a semi-supervised manner. To make student networks generalize better than the teacher network, noise is injected during training of student networks. According to evaluation results our proposed model achieves state-of-the-art accuracy in monocular depth estimation.


Feedback Motion Planning For a Dynamic Car Model via Random Sequential Composition
Özcan, Melih; Ankaralı, Mustafa Mert (2019-01-01)
Autonomous cars and car-like robots have gained huge popularity recently due to the recent advancements in technology and AI industry. Motion and path planning is one of the most fundamental problems for such systems. In the literature, kinematic models are widely adopted for planning and control for these type of robots due to their simplicity (control and analysis) and fewer computational requirements. Though, applicability of kinematic models are limited to very low speeds or some specific cases, which c...
3D imaging via binary wavefront modulation for lidar and machine vision applications
Yüksel, Çağdaş Anıl; Yüce, Emre; Department of Micro and Nanotechnology (2020-11-2)
Autonomous vehicles have proven to be very efficient in daily routine jobs and their impact will continue to increase given the recent developments in artificial intelligence, boosted by increased computation capacity. These vehicles are generally equipped with 2D imaging sensors and asked to accomplish tasks in a 3D world, which hamper their functionalities. In this study, we experimentally investigate and develop 3D imaging technologies. We first demonstrate colorful 3D imaging via time of flight me...
Fully autonomous mini/micro scale UAV field experiences and image processing applications Tam Otonom Mini/Mikro Ölçekli IHA Saha Deneyimleri ve Görüntü Isleme Uygulamalari
Bil, Dilek Basaran; Konukseven, Erhan İlhan (2018-07-05)
Fully autonomous mini/micro scale rotary and fixed wing UAV R&D works, while increasing their environmental and conditional awareness with the image processing techniques with some different results and experiences will shared within this paper. Field test experiences and lessons learned on autonomous mobile mini/micro scale UAV systems, some results and open fields on image processing techniques tried to be illustrated on the obtained image results. Finally critical and important points on image processing...
Characterization of Driver Neuromuscular Dynamics for Human-Automation Collaboration Design of Automated Vehicles
Lv, Chen; Wang, Huaji; Cao, Dongpu; Zhao, Yifan; Auger, Daniel J.; Sullman, Mark; Matthias, Rebecca; Skrypchuk, Lee; Mouzakitis, Alexandros (Institute of Electrical and Electronics Engineers (IEEE), 2018-12-01)
In order to design an advanced human-automation collaboration system for highly automated vehicles, research into the driver's neuromuscular dynamics is needed. In this paper, a dynamic model of drivers' neuromuscular interaction with a steering wheel is first established. The transfer function and the natural frequency of the systems are analyzed. In order to identify the key parameters of the driver-steering-wheel interacting system and investigate the system properties under different situations, experim...
Fuzzy Decision Fusion for Single Target Classification in Wireless Sensor Networks
Gok, Sercan; Yazıcı, Adnan; Coşar, Ahmet; George, Roy (2010-07-23)
With the advances in technology, low cost and low footprint sensors are being used more and more commonly. Especially for military applications wireless sensor networks (WSN) have become an attractive solution as they have great use for avoiding deadly danger in combat. For military applications, classification of a target in a battlefield plays an important role. A wireless sensor node has the ability to sense the raw signal data in battlefield, extract the feature vectors from sensed signal and produce a ...
Citation Formats
C. B. Süvari, “Semi-supervised iterative teacher-student learning for monocular depth estimation,” M.S. - Master of Science, Middle East Technical University, 2021.