Generation of synthetic training data for object detection in piles

Buls, Elvijs; Kadiķis, Roberts; Cacurs, Ričards; Ārents, Jānis

doi:10.1117/12.2523203

Cited by 6 publications

(5 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Synthetic data for object detection tasks typically has been composed by placing foreground objects on background scenes with different parameters that can be varied. Some approaches proceed with 2D images that are placed on a set of background images [7] [11], with set of rules or physical simulation for piling the objects realistically. The level of complexity for such and similar methods is lower, however they are typically restricted to 2D nature that contributes to the lack of realism, which therefore decreases detectors performance.…”

Section: Related Workmentioning

confidence: 99%

Synthetic Data of Randomly Piled, Similar Objects for Deep Learning-Based Object Detection

Arents,

Lesser,

Bizuns

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Currently, the best object detection results are achieved by supervised deep learning methods, however, these methods depend on annotated training data. With the synthetic data generation approach, we intend to mimic the real data characteristics and diversify the dataset by a systematic rendering of highly realistic synthetic pictures. We systematically explore how different combinations and portions of real and synthetic datasets affect object detectors performance. The developed synthetic data generation framework shows promising results in deep learning-based object detection tasks and can supplement real data when the variety of real training data is insufficient. However, when synthetic data ratio increases over real data ratio, a decrease in average precision can be observed, which has the most affect on 0.75-0.95 IoU threshold range.

show abstract

Section: Related Workmentioning

confidence: 99%

Synthetic Data of Randomly Piled, Similar Objects for Deep Learning-Based Object Detection

Arents,

Lesser,

Bizuns

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

show abstract

“…Inception v3 [570], ResNet [238], and Xception [117] by fine-tuning them on the ADORESet dataset that contains 2500 real and 750 synthetic images for each of 30 object categories in the context of robotic manipulation; they find that a hybrid dataset achieves much better recognition quality compared to purely synthetic or purely real datasets. Recent applications of synthetic data for object detection include the detection of objects in vending machines [623], objects in piles for training robotic arms [75], computer game objects [560], smoke detection [663], deformable part models [681], face detection in biomedical literature [140], drone detection [504], and more. Recently, Nowruzi et al [426] studied the question of how much real data is actually needed for object detection in comparison to synthetic data.…”

Section: Improving High-level Computer Vision With Synthetic Datamentioning

confidence: 99%

Synthetic Data for Deep Learning

Nikolenko¹

2019

Preprint

View full text Add to dashboard Cite

Synthetic data is an increasingly popular tool for training deep learning models, especially in computer vision but also in other areas. In this work, we attempt to provide a comprehensive survey of the various directions in the development and application of synthetic data. First, we discuss synthetic datasets for basic computer vision problems, both low-level (e.g., optical flow estimation) and high-level (e.g., semantic segmentation), synthetic environments and datasets for outdoor and urban scenes (autonomous driving), indoor scenes (indoor navigation), aerial navigation, simulation environments for robotics, applications of synthetic data outside computer vision (in neural programming, bioinformatics, NLP, and more); we also survey the work on improving synthetic data development and alternative ways to produce it such as GANs. Second, we discuss in detail the synthetic-to-real domain adaptation problem that inevitably arises in applications of synthetic data, including syntheticto-real refinement with GAN-based models and domain adaptation at the feature/model level without explicit data transformations. Third, we turn to privacy-related applications of synthetic data and review the work on generating synthetic datasets with differential privacy guarantees. We conclude by highlighting the most promising directions for further work in synthetic data studies.

show abstract

“…Recent approaches in using synthetic training data proved to be rather successful [51][52][53][54][55][56]. In particular, the creation of synthetic images by domain randomization using CAD models in [53][54][55][56] seemed superior compared to composition [51,52]. An illustration of the methods is shown in Figure 2.6.…”

Section: Introduction Of Synthetic Data For Deep Learning Methodsmentioning

confidence: 99%

“…Generation of synthetic images using composition does not require the CAD model of the object. E. Buls et al [51] worked on generating a pile of objects using composition by capturing different views of real objects and stitching them together, however, the detection results were not favorable as the accuracy of correctly detected objects ranged from 10% to 29% for different objects.…”

Section: Synthetic Rendering By Compositionmentioning

confidence: 99%

Robotic part manipulation for 3D-print post-processing

Lim

View full text Add to dashboard Cite

The contributions of the co-authors are as follows:I wish to express my greatest gratitude to my spouse, family, and friends who have given me their utmost encouragement and support throughout this journey. My heartfelt appreciation also goes to my NTU supervisor and both HP managers for their guidance. I am also very grateful to my fellow colleagues, my thesis advisory committee, and my examiners for their directions.

show abstract

Generation of synthetic training data for object detection in piles

Cited by 6 publications

References 9 publications

Synthetic Data of Randomly Piled, Similar Objects for Deep Learning-Based Object Detection

Synthetic Data of Randomly Piled, Similar Objects for Deep Learning-Based Object Detection

Synthetic Data for Deep Learning

Robotic part manipulation for 3D-print post-processing

Contact Info

Product

Resources

About