1,721,060 research outputs found
CAD2Render: A Modular Toolkit for GPU-accelerated Photorealistic Synthetic Data Generation for the Manufacturing Industry
No full textNo description provided.PILS SBO: Product Inspectie with Little Supervision. Flanders Make (Belgium). awardNumber:null. 02ndjfz59NORM.AI SBO. Flanders Make (Belgium). awardNumber:null. 02ndjfz5
EDM-Research/UE-LASAA-ALVR: v1.0
No full textUnreal Engine 5.3 plugin for Large-area Spatially Aligned Anchors using ALVR remote renderingMAX-R. European Europe Project. awardNumber:101070072
Leveraging Transfer Learning for Niche Sign System Recognition in VR Training with Limited Data
No full textSpeech Supported by Gestures, in Dutch Spreken met On-dersteuning van Gebaren (SMOG), is a Belgian sign system that enhances verbal communication for individuals with communicative disabilities through specific gestures. Although effective, SMOG training is labor-intensive and typically requires one-on-one instruction. Virtual reality (VR) training can make the practice of SMOG gestures more scalable, playful, and cost-effective. For such a VR training to be effective , trainees should receive accurate and timely automated feedback on whether they perform the correct gestures. However, since SMOG and many other specialized motor skills are niche problems, there are no large annotated datasets to train machine learning models to perform this task from scratch, and collecting large amounts of data for such niche tasks is infeasible. We therefore propose using transfer learning to fine-tune pre-trained Mobile Video Networks (MoViNets) on a small dataset of RGB videos showing SMOG gestures. Through this workflow, we demonstrate recognition accuracies exceeding 99% using only two to five samples per gesture for training. This work therefore not only advances accessible SMOG training through autonomous VR practice but also establishes a highly data-and computation-efficient machine-learning framework for recognizing other niche sign systems or motor skills using limited amounts of training data.We thank Yasmine Wauthier from the Expertise Center for Lifelong Learning and Innovation (OLLI), the Orthopedagogical Guidance Graduate Program of the AP University of Applied Arts and Sciences, and SMOG vzw for their support. This work was made possible with the support of MAXVR-INFRA, a scalable and flexible infrastructure that facilitates the transition to digital-physical work environments
Genetic Learning for Designing Sim-to-Real Data Augmentations Source Code
No full textWe analyze the benefit of data augmentations for overcoming the sim-to-real gap and use the results to develop a genetic learning algorithm for finding augmentation policies
Genetic Learning for Designing Sim-to-Real Data Augmentations
No full textData augmentations are useful in closing the sim-to-real domain gap when training on synthetic data. This is because they widen the training data distribution, thus encouraging the model to generalize better to other domains. Many image augmentation techniques exist, parametrized by different settings, such as strength and probability. This leads to a large space of different possible augmentation policies. Some policies work better than others for overcoming the sim-to-real gap for specific datasets, and it is unclear why. This paper presents two different interpretable metrics that can be combined to predict how well a certain augmentation policy will work for a specific sim-to-real setting, focusing on object detection. We validate our metrics by training many models with different augmentation policies and showing a strong correlation with performance on real data. Additionally, we introduce GeneticAugment, a genetic programming method that can leverage these metrics to automatically design an augmentation policy for a specific dataset without needing to train a model
Projector-camera calibration with non-overlapping fields of view using a planar mirror
No full textProjector-camera systems have numerous applications across diverse domains. Accurate calibration of both intrinsic and extrinsic parameters is crucial for these systems. Intrinsic parameters include focal length, distortion parameters, and the principal point, while extrinsic parameters encompass the position and orientation of the projector and camera. A non-overlapping projector-camera system is required in certain scenarios due to practical limitations, physical arrangements, or specific application requirements. These systems pose a more complex calibration challenge because the devices have no direct correspondences or overlapping fields of view, necessitating intermediate objects or methods. This paper proposes a calibration method for non-overlapping projector-camera systems using a planar mirror. The method involves a straightforward process that requires a calibrated camera and a separate mirror calibration step. In this setup, the projector displays a pattern on a planar calibration board, and the camera has an indirect view of this calibration board through the mirror. Using homography, 3D-2D correspondences of the projector are established, enabling the calibration of the system. This method is empirically evaluated using real-world setups and quantitatively assessed in a synthetic environment. The results demonstrate that the proposed method achieves precise calibration across various setups, proving its effectiveness. This approach is an easy-to-use and accessible calibration process for non-overlapping projector-camera systems, made possible by a mirror.This research was partly funded by the European Union
(HORIZON MAX-R, Mixed Augmented and Extended Reality Media
Pipeline, 101070072), the Flanders Make’s XRTwin SBO project
(R-12528), the Special Research Fund (BOF) of Hasselt University
(R-14360) and the specialized FWO fellowship grant (1SHDZ24N).
Data availability The datasets generated during and/or analyzed during
the current study are available from the corresponding author on
reasonable request
Active learning for quality inspecting with synthetic hot- start approach
No full textIn the pharmaceutical industry, there are strict requirements on the presence of contaminants inside single-use syringes (so-called unijects). Quality management systems include various methods such as measuring weight, manual inspection or vision techniques. Automated and accurate techniques for quality inspection are preferred, reducing the costs and increasing the speed of production.
In this paper we analyze defects on unijects. During inspection, the product is spun around to force contaminants to the outside of the bulb and photos are taken. These photos can be manually inspected, however using computer vision techniques this process can be automated.
As such inclusions are exceedingly rare to occur in practice, it is very difficult to collect a first dataset to train a deep-learning network on, which contains actual defects. The approach we will demonstrate in our contribution introduces synthetic defects on top of regular images for kickstarting the defect detection network. Using this initial defect segmentation network, we can then introduce classic uncertainty and diversity sampling algorithms to select relevant images for annotation. Normally, in these 'active learning' strategies the initial dataset is taken at random. However, because of the low probability of selecting each type of defect at random, the model has a very cold start. We will demonstrate how our hot-start approach using synthetic defects solves this initialization problem
Analysis of Training Object Detection Models with Synthetic Data
No full textRecently, the use of synthetic training data has been on the rise as it offers correctly labelled datasets at a lower cost. The downside of this technique is that the so-called domain gap between the real target images and synthetic training data leads to a decrease in performance. In this paper, we attempt to provide a holistic overview of how to use synthetic data for object detection. We analyse aspects of generating the data as well as techniques used to train the models. We do so by devising a number of experiments, training models on the Dataset of Industrial Metal Objects (DIMO). This dataset contains both real and synthetic images. The synthetic part has different subsets that are either exact synthetic copies of the real data or are copies with certain aspects randomised. This allows us to analyse what types of variation are good for synthetic training data and which aspects should be modelled to closely match the target data. Furthermore, we investigate what types of training techniques are beneficial towards generalisation to real data, and how to use them. Additionally, we analyse how real images can be leveraged when training on synthetic images. All these experiments are validated on real data and benchmarked to models trained on real data. The results offer a number of interesting takeaways that can serve as basic guidelines for using synthetic data for object detection. Code to reproduce results is available at https://github.com/EDM-Research/DIMO_ObjectDetection
EDM-Research/DIMO_ObjectDetection: v1.0
No full textObject detection for the DIMO dataset. Uses the Mask-RCNN model. This is the official implementation of Analysis of Training Object Detection Models with Synthetic Data, published in BMVC: British Machine Vision Conference, 2022.
Source code for the following scientific publication:
Vanherle, B., Moonen, S., Van Reeth, F., and Michiels, N. (2022). Analysis of Training Object Detection Models with Synthetic Data. 33rd British Machine Vision Conference 2022, BMVC 2022, London, UK, November 21-24, 2022. Retrieved from https://bmvc2022.mpi-inf.mpg.de/0833.pdfPILS SBO: Product Inspection with Little Supervision. Flanders Make (Belgium). awardNumber:null. 02ndjfz59BOF Special Research Fund. Hasselt University. awardNumber:null. 10.13039/50110000955
- …
