2024
2024
Bridging the Reality Gap in Industrial Synthetic Datasets with Aging Texture Simulation
Joe Khalil*, Jimmy Tekli, Angelo Yaghi, Marc Kamradt, Najib Metni, Raphael Couturier
As industries increasingly rely on computer vision applications, the demand for robust object detection models grows. Synthetic datasets offer a promising solution by providing diverse and accurate data scenarios, enhancing the performance of these models. In this study, we tackle one aspect of the ongoing challenge to bridge the reality gap between synthetic and real images in industrial applications. We incorporated textures that accurately replicate signs of usage on industrial assets. Furthermore, we conducted a comprehensive experiment featuring three distinct datasets from the same simulated environment to evaluate the impact of texture variations on model performance. Our approach introduces a combination of aging texture layers to six industrial assets for object detection. To assess performance, we trained three Deep Learning architectures with these datasets and evaluated them by inferring on real images captured in industrial settings. The results of our experiments demonstrate the effectiveness of the aging textures in real-world scenarios, affirming the success of our proposed approach in enhancing object detection models for industrial applications.
-
(Coming Soon)
-
(Coming Soon)
Synthetic Data: Revolutionizing the Industrial Metaverse
Jimmy Nassif, Joe Tekli, Marc Kamradt
The book concentrates on the impact of digitalization and digital transformation technologies on the Industry 4.0 and smart factories, how the factory of tomorrow can be designed, built, and run virtually as a digital twin likeness of its real-world counterpart, before the physical structure is actually erected.
It highlights the main digitalization technologies that have stimulated the Industry 4.0, how these technologies work and integrate with each other, and how they are shaping the industry of the future.
It examines how multimedia data and digital images in particular are being leveraged to create fully virtualized worlds in the form of digital twin factories and fully virtualized industrial assets. It uses BMW Group’s latest SORDI dataset (Synthetic Object Recognition Dataset for Industry), i.e., the largest industrial images dataset to-date and its applications at BMW Group and Idealworks, as one of the main explanatory scenarios throughout the book.
It discusses the need of synthetic data to train advanced deep learning computer vision models, and how such datasets will help create the “robot gym” of the future: training robots on synthetic images to prepare them to function in the real world.
-
Nassif, Jimmy, Joe Tekli, and Marc Kamradt. Synthetic Data: Revolutionizing the Industrial Metaverse. Springer Nature, 2024.
-
@book{nassif2024synthetic,
title={Synthetic Data: Revolutionizing the Industrial Metaverse},
author={Nassif, Jimmy and Tekli, Joe and Kamradt, Marc},
year={2024},
publisher={Springer Nature}
}
2023
2023
Enhancing Complex Image Synthesis with Conditional Generative Models and Rule Extraction
Chafic ABOU AKAR*, Andre Luckow, Ahmad Obeid, Christian Beddawi, Marc KAMRADT, Abdallah MAKHOUL
Generative Adversarial Networks (GANs) have shown potential for generating images, but have limitations when applied to complex datasets. To address these limitations, class-conditional training is employed, as it performs better and maintains a high level of semantic diversity. In this work, we propose a new method for training generative models on complex images by extracting rules defining the relationships between objects in the image, cropping significant sub-regions based on these rules, and training the models in a conditional setting using the extracted rules as labels. The proposed approach is evaluated, and the results demonstrate its effectiveness by increasing the training dataset size, and then feeding it to conditional training. As a result, synthesized samples maintain asset fine-grained details and the visibility of small instances.
-
Abou Akar, Chafic, et al. "Enhancing Complex Image Synthesis with Conditional Generative Models and Rule Extraction." 2023 International Conference on Machine Learning and Applications (ICMLA). IEEE, 2023.
-
@inproceedings{abou2023enhancing,
title={Enhancing Complex Image Synthesis with Conditional Generative Models and Rule Extraction},
author={Abou Akar, Chafic and Luckow, Andre and Obeid, Ahmad and Beddawi, Christian and Kamradt, Marc and Makhoul, Abdallah},
booktitle={2023 International Conference on Machine Learning and Applications (ICMLA)},
pages={136--143},
year={2023},
organization={IEEE}
}
Mixing Domains for Smartly Picking and Using Limited Datasets in Industrial Object Detection
Chafic ABOU AKAR*, Anthony SEMAAN, Youssef HADDAD, Marc KAMRADT, Abdallah MAKHOUL
Object detection is a popular computer vision task that is performed by autonomous industrial robots. However, training a detection model requires a large annotated image dataset that belongs to the camera domain of the robot (the test domain). Acquiring such data in a similar domain or rendering photo-realistic images from a realistic virtual environment composed of accurate 3D models and using powerful hardware, can be expensive, time-consuming, and requires specialized expertise. This article focuses on investigating the growth of average precision (AP) in object detection as we progressively train and test our models using various combinations of acquired and rendered datasets from different domains: real and synthetic. By analyzing the results on industrial load carrier box detection, we discovered that a hybrid dataset comprising 20–30% of images similar to the test domain leads to achieving nearly maximum detection accuracy.
-
Abou Akar, C., Semaan, A., Haddad, Y., Kamradt, M., Makhoul, A. (2023). Mixing Domains for Smartly Picking and Using Limited Datasets in Industrial Object Detection. In: Christensen, H.I., Corke, P., Detry, R., Weibel, JB., Vincze, M. (eds) Computer Vision Systems. ICVS 2023. Lecture Notes in Computer Science, vol 14253. Springer, Cham. https://doi.org/10.1007/978-3-031-44137-0_23
-
@inproceedings{abou2023mixing,
title={Mixing Domains for Smartly Picking and Using Limited Datasets in Industrial Object Detection},
author={Abou Akar, Chafic and Semaan, Anthony and Haddad, Youssef and Kamradt, Marc and Makhoul, Abdallah},
booktitle={International Conference on Computer Vision Systems},
pages={270--282},
year={2023},
organization={Springer}
}
2022
2022
Synthetic Object Recognition Dataset for Industries
Chafic ABOU AKAR*, Jimmy TEKLI+, Daniel JESS+, Mario KHOURY, Marc KAMRADT, Michael GUTHE
Smart robots in factories highly depend on Computer Vision (CV) tasks, e.g. object detection and recognition, to perceive their surroundings and react accordingly. These CV tasks can be performed after training deep learning (DL) models on large annotated datasets. In an industrial setting, acquiring and annotating such datasets is challenging because it is time-consuming, prone to human error, and limited by several privacy and security regulations. In this study, we propose a synthetic industrial dataset for object detection purposes created using NVIDIA Omniverse. The dataset consists of 8 industrial assets in 32 scenarios and 200,000 photo-realistic rendered images that are annotated with accurate bounding boxes. For evaluation purposes, multiple object detectors were trained with synthetic data to infer on real images captured inside a factory. Accuracy values higher than 50% and up to 100% were reported for most of the considered assets.
-
C. A. Akar, J. Tekli, D. Jess, M. Khoury, M. Kamradt and M. Guthe, "Synthetic Object Recognition Dataset for Industries," 2022 35th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI), 2022, pp. 150-155, doi: 10.1109/SIBGRAPI55357.2022.9991784.
-
@INPROCEEDINGS{9991784,
author={Akar, Chafic Abou and Tekli, Jimmy and Jess, Daniel and Khoury, Mario and Kamradt, Marc and Guthe, Michael},
booktitle={2022 35th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI)},
title={Synthetic Object Recognition Dataset for Industries},
year={2022},
volume={1},
number={},
pages={150-155},
doi={10.1109/SIBGRAPI55357.2022.9991784}}