Training Deep Neural Networks for Detecting Drinking Glasses Using Synthetic Images

Jabbar, Abdul; Farrawell, Luke; Fountain, Jake; Chalup, Stephan K.

doi:10.1007/978-3-319-70096-0_37

Cited by 22 publications

(17 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This value is higher than the threshold of around 15 fps, which is a good frame rate for video (Chen et al, 2007). The IoU was about 0.77, which is higher than 0.5, the value of IoU at which buildings were adequately detected (Jabbar et al, 2017).…”

Section: Discussionmentioning

confidence: 71%

How a Flooded City Can Be Visualized from Both the Air and the Ground with the City Digital Twin Approach, System Integration of Flood Simulation and Augmented Reality with Drones

Kikuchi¹,

Fukuda²,

Yabuki³

2022

CAADRIA Proceedings

View full text Add to dashboard Cite

City digital twins are becoming increasingly important for the sustainable development of cities, and augmented reality (AR) has been attracting attention as a tool for visualizing city digital twins. In addition, from the perspective of SDG 11, it is essential to manage flood risk in urban spaces. However, there are no case studies that present a bird's-eye view of a simulated city. Visualizing the state of a flooded city during a disaster is one potential use case. From the perspective of information graphics, people want to understand urban data at the micro and macro levels. This study proposes a city-digital-twin approach for visualizing a simulated city using a large-scale AR and drone integration method that does not require a specific software development kit (SDK). This system can visualize the state of a city flooded by a disaster from both a bird's-eye view of the city at several tens of metres above it and from a first-person perspective of the user's area of activity. The applicability of the system is demonstrated through verification and case studies.

show abstract

Section: Discussionmentioning

confidence: 71%

How a Flooded City Can Be Visualized from Both the Air and the Ground with the City Digital Twin Approach, System Integration of Flood Simulation and Augmented Reality with Drones

Kikuchi¹,

Fukuda²,

Yabuki³

2022

CAADRIA Proceedings

View full text Add to dashboard Cite

show abstract

“…Blender uses a path tracer rendering engine to generate physics-based renderings and can be fully automated using Python scripts. For those reasons, Blender is a popular tool amongst researchers for the automated generation of training images for deep learning (e.g., [ 30 , 32 , 34 , 44 , 45 ]). Our workflow and the scope of investigated aspects are depicted in Figure 1 , given the constraint that the image generation pipeline can be easily adapted to new objects and industrial use cases.…”

Section: Methodsmentioning

confidence: 99%

“…Jabbar et al [ 32 ] used the software Blender to create photorealistic images of transparent drinking glasses. They used high dynamic range images (HDRIs) as 360 degree background images, which also provide complex image-based lighting (IBL) [ 33 ], thus removing the need to manually setup realistic lighting in the scene.…”

Section: Related Workmentioning

confidence: 99%

Generating Images with Physics-Based Rendering for an Industrial Object Detection Task: Realism versus Domain Randomization

Eversberg

Lambrecht

2021

Sensors

View full text Add to dashboard Cite

Limited training data is one of the biggest challenges in the industrial application of deep learning. Generating synthetic training images is a promising solution in computer vision; however, minimizing the domain gap between synthetic and real-world images remains a problem. Therefore, based on a real-world application, we explored the generation of images with physics-based rendering for an industrial object detection task. Setting up the render engine’s environment requires a lot of choices and parameters. One fundamental question is whether to apply the concept of domain randomization or use domain knowledge to try and achieve photorealism. To answer this question, we compared different strategies for setting up lighting, background, object texture, additional foreground objects and bounding box computation in a data-centric approach. We compared the resulting average precision from generated images with different levels of realism and variability. In conclusion, we found that domain randomization is a viable strategy for the detection of industrial objects. However, domain knowledge can be used for object-related aspects to improve detection performance. Based on our results, we provide guidelines and an open-source tool for the generation of synthetic images for new industrial applications.

show abstract

“…Deep learning comes with challenges with respect to computational resources and training data requirements [6,13]. Some of the breakthroughs in deep neural networks (DNNs) only became possible through the availability of massive computing systems or through careful co-design of software and hardware.…”

Section: Introductionmentioning

confidence: 99%

Comparing Computing Platforms for Deep Learning on a Humanoid Robot

Biddulph

Houliston

Mendes

et al. 2018

Neural Information Processing

Self Cite

View full text Add to dashboard Cite

The goal of this study is to test two different computing platforms with respect to their suitability for running deep networks as part of a humanoid robot software system. One of the platforms is the CPUcentered Intel R NUC7i7BNH and the other is a NVIDIA R Jetson TX2 system that puts more emphasis on GPU processing. The experiments addressed a number of benchmarking tasks including pedestrian detection using deep neural networks. Some of the results were unexpected but demonstrate that platforms exhibit both advantages and disadvantages when taking computational performance and electrical power requirements of such a system into account.

show abstract

Training Deep Neural Networks for Detecting Drinking Glasses Using Synthetic Images

Cited by 22 publications

References 8 publications

How a Flooded City Can Be Visualized from Both the Air and the Ground with the City Digital Twin Approach, System Integration of Flood Simulation and Augmented Reality with Drones

How a Flooded City Can Be Visualized from Both the Air and the Ground with the City Digital Twin Approach, System Integration of Flood Simulation and Augmented Reality with Drones

Generating Images with Physics-Based Rendering for an Industrial Object Detection Task: Realism versus Domain Randomization

Comparing Computing Platforms for Deep Learning on a Humanoid Robot

Contact Info

Product

Resources

About