Machine Graphics and Vision

An efficient pedestrian attributes recognition system under challenging conditions

Ha X. Nguyen, Dong N. Hoang, Tuan A. Tran, Tuan M. Dang — Mon, 21 Aug 2023 00:00:00 +0000

In this work, an efficient pedestrian attribute recognition system is introduced. The system is based on a novel processing pipeline that combines the best-performing attribute extraction model with an efficient attribute filtering algorithm using keypoints of human pose. The attribute extraction models are developed based on several state-of-the-art deep networks via transfer learning techniques, including ResNet50, Swin-transformer, and ConvNeXt. Pre-trained models of these networks are fine-tuned using the Ensemble Pedestrian Attribute Recognition (EPAR) dataset. Several optimization techniques, including the advanced optimizer Adam with Decoupled Weight Decay Regularization (AdamW), Random Erasing (RE), and weighted loss functions, are adopted to solve issues of data unbalancing or challenging conditions like partial and occluded bodies. Experimental evaluations are performed via EPAR that contains 26993 images of 1477 person IDs, most of which are in challenging conditions. The results show that the ConvNeXt-v2-B outperforms other networks; mean accuracy (mA) reaches 85.57%, and other indices are also the highest. The addition of AdamW or RE can improve accuracy by 1-2%. The use of new loss functions can solve the issue of data unbalancing, in which the accuracy of data-less attributes improves by a maximum of 14% in the best case. Significantly, when the attribute filtering algorithm is applied, the results are dramatically improved, and mA reaches an excellent value of 94.85%. Utilizing the state-of-the-art attribute extraction model with optimization techniques on the large-scale and diverse dataset and attribute filtering has shown a good approach and thus has a high potential for practical applications.

Use of virtual reality to facilitate engineer training in the aerospace industry

Mon, 06 Nov 2023 00:00:00 +0000

This work concerns automation of the training process, using modern information technologies, including virtual reality (VR). The starting point is an observation that automotive and aerospace industries require effective methods of preparation of engineering personnel. In this context, the technological process of preparing operations of a CNC numerical machine has been extracted. On this basis, a dedicated virtual reality environment, simulating manufacturing of a selected aircraft landing gear component, was created. For a comprehensive analysis of the pros and cons of the proposed approach, four forms of training, involving a physical CNC machine, a physical simulator, a software simulator, and the developed VR environment were instantiated. The features of each training form were analysed in terms of their potential for industrial applications. A survey, using the Net Promoter Score method, was also conducted among a target group of engineers, regarding the potential of use of each training form. As a result, the advantages and disadvantages of all four training forms were captured. They can be used as criteria for selecting the most effective training form.

Generating layout for complex cave-like levels with schematic maps and Cellular Automata

Izabella Antoniuk — Mon, 11 Dec 2023 00:00:00 +0000

In this paper an algorithm for creating cave-like, user-guided layout is presented. In applications such as computer games, underground structures offer unique challenges and interesting space for player actions. Preparation of such areas can be time consuming and tiresome, especially during the design process, when many ideas are often scrapped. Presented approach aims at improving this process. Schematic input is used so the user can quickly define the general layout. Cave system is divided into levels and tiles - easily-parallelizable modules for the following method stages. Cellular automata are used to extend initial system sketch with interesting shapes while the diamond-square algorithm spreads the final terrain heights. Each stage uses the results of the previously performed operations as input, providing space for alterations. Input maps can be reused to obtain different variations of the same system. The final structure is represented as a 3D point cloud. Chosen representation supports multilevel systems and can be used either as a base for further algorithms, or as a final mesh. The presented approach can be easily incorporated into game design process, while visualizing initial layouts and speeding up preparation of unique, interesting and challenging game spaces for the players to traverse.

Assessment of the possibility of imitating experts' aesthetic judgments about the impact of knots on the attractiveness of furniture fronts made of pine wood

Krzysztof Gajowniczek, Marcin Bator, Katarzyna Śmietańska, Jarosław Górski — Tue, 12 Dec 2023 00:00:00 +0000

Our research aims to reconstruct expert preferences regarding the visual attractiveness of furniture fronts made of pine wood using machine learning algorithms. A numerical experiment was performed using five machine learning algorithms of various paradigms. To find the answer to the question of what determines the expert's decision, we determined the importance of variables for some machine learning models. For random forest and classification trees, it involves the overall reduction in node impurities resulting from variable splitting, while for neural networks it uses the Garson algorithm. Based on the numerical experiments we can conclude that the best results of expert decision reconstruction are provided by a neural network model. The expert's decision is better reconstructed for more beautiful images. The decision for nice images is made based on the best 4 or 5 variables, while for ugly images many more features are important. Prettier images and those for which the expert's decision is better reconstructed have fewer knots.

Advancing chipboard milling process monitoring through spectrogram-based time series analysis with Convolutional Neural Network using pretrained networks

Jarosław Kurek, Karol Szymanowski, Leszek Chmielewski, Arkadiusz Orłowski — Tue, 12 Dec 2023 00:00:00 +0000

This paper presents a novel approach to enhance chipboard milling process monitoring in the furniture manufacturing sector using Convolutional Neural Networks (CNNs) with pretrained architectures like VGG16, VGG19, and RESNET34. The study leverages spectrogram representations of time-series data obtained during the milling process, providing a unique perspective on tool condition monitoring. The efficiency of the CNN models in accurately classifying tool conditions into distinct states (`Green', `Yellow', and `Red') based on wear levels is thoroughly evaluated. Experimental results demonstrate that VGG16 and VGG19 achieve high accuracy, however with longer training times, while RESNET34 offers faster training at the cost of reduced precision. This research not only highlights the potential of pretrained CNNs in industrial applications but also opens new avenues for predictive maintenance and quality control in manufacturing, underscoring the broader applicability of AI in industrial automation and monitoring systems.