Machine Graphics and Vision

Verification of data compression focusing on continuity in 3D printing

Satoshi Kodama — 2024-12-23

Recently, 3D printers have become capable of producing relatively large, high-resolution models. Unlike simple shapes, it is becoming possible to print large complex shapes with high accuracy. However, the data size of complex models is also large, and the slice data required for printing is also large. Thus, in this study, we investigated reducing the data size by focusing on the characteristics of the slice data required for 3D printing. The proposed method focuses on the continuity of each layer and the top/bottom layers of the cross-section used to print the 3D model. Preliminary experiments were conducted to determine whether the data size could be reduced by applying the difference method. Here, the results obtained from the continuity were output as text data, and various metadata, e.g., lamination pitch data, required for printing were ZIP compressed. Then, we compared conventional file formats as a format that can be converted as a printable file as lossless compression. The results demonstrated that the file size can be reduced for 3D complex shapes with a large number of vertices, which are difficult to handle. We found that the proposed difference method was effective for relatively large files that require a general-purpose graphics processing unit to create slice data.

Determination of spherical coordinates of sampled cosmic ray flux distribution using Principal Components Analysis and deep Encoder-Decoder network

Tomasz Hachaj — 2024-12-23

In this paper we propose a novel algorithm based on the use of Principal Components Analysis for the determination of spherical coordinates of sampled cosmic ray flux distribution. We have also applied a deep neural network with encoder-decoder (E-D) architecture in order to filter-off variance noises introduced by sampling. We conducted a series of experiments testing the effectiveness of our estimations. The training set consisted of 92250 images and validation set of 37800 images. We have calculated mean absolute error (MAE) between real values and estimations. When E-D is applied, the number of cases (estimations) where MAE < 10 increases from 48% to 79% for θ and from 62% to 65% for ϕ, MAE < 5 increases from 24% to 45% for θ and from 47% to 52% for ϕ, MAE < 1 increases from 6% to 9% for θ and from 12% to 16% for ϕ, where θ is the zenith angle, and ϕ is the azimuthal angle. This is a significant change and it demonstrates the high utility of the E-D network use and shows the accuracy of the PCA-based algorithm. We also publish the source code used in our research in order to make it reproducible.

A comparative study of DeepLabCut and other open-source pupillometry data analysis algorithms – Which to choose?

Amitesh Badkul — 2024-12-23

Pupillometry measures pupil size, and several open-source algorithms are available to analyse pupillometry data. However, only a few studies compared these algorithms' accuracy and computational resources. This study aims to compare the accuracy of computer vision-based algorithms (Swirski, Starburst, PuRe, ElSe, ExCuSe algorithms) and the machine learning algorithm, DeepLabCut, to the double-blinded human examiners (gold-standard). Training of DeepLabCut with different architectures and a variable number of markers (2-9 markers) was done on an open-source dataset. The duration of training was statistically longer for the ResNet152 model compared to the MobileNet model. The pupil diameters in computer vision-based software such as PuRe, Starburst, and Swirski were statistically different from human measurements. MobileNet 2 and 3 marker models were the closest to the human measurements. In conclusion, this work highlights the efficiency of lower marker models based on MobileNet architecture in DeepLabCut, which consumes fewer computational resources and is more accurate.

Machine vision for automated maturity grading of oil palm fruits: A systematic review

Afsar Kamal — 2024-12-23

The maturity of oil palm fruits is a very crucial factor for oil extraction industry in Indonesia, Malaysia, Thailand, and other countries to ensure the oil quality and increase productivity. This literature review examines the various machine learning techniques, especially the deep learning techniques used to automate the maturity grading process of oil palm fresh fruit bunches. The crucial advantages of using machine learning approaches were highlighted, and the limitations and prospects of each research article were discussed. This review describes the various image pre-processing techniques utilized to prepare images for model training. CNN is identified as the dominant over all classification techniques of machine learning to classify the oil palm fruits images based on maturity level, due to its ability of learning complex features.