A procedure for the determination of the location of prostate implant seeds that are wholly overlapped in a projection view has been developed. The procedure mainly consists of a series of image processing and an in-house developed localization software based on a three-film technique. To verify the efficacy of the procedure, a simulation phantom was built and nine sets of simulation were performed. For the assessment of the location of the seeds in the phantom, three images, one in anterior-posterior direction and two others in oblique angles, were acquired and a series of image processing was applied to the images for the removal of unnecessary background and the improvement of imaging quality. In this study, three types were considered; first, when two seeds were overlapped in one of projection images, second, more than three seeds were overlapped in one of projection images, and the third, all images contained wholly overlapped seeds. The developed software separates wholly overlapped seeds by calculating the distance between seeds in each film. This software can provide valuable information for establishing effective quality assurance in permanent prostate brachytherapy.

Mobile cranes are essential for transporting heavy materials at construction sites, but their operation carries significant safety risks, particularly due to the potential for overturning accidents. These accidents can be classified into two main categories: mechanical accidents, which are caused by factors such as outrigger failure, excessive load weight, and operator skill, and environmental accidents, which arise from ground subsidence due to groundwater and sinkholes. While numerous studies have addressed the causes and prevention of mechanical accidents, there has been a lack of research focusing on the prevention of environmental accidents. This study presents the development of an Electrical Resistivity Measurement System (ERMS) designed to prevent overturning accidents caused by ground subsidence at mobile crane work sites. The ERMS, mounted on a mobile crane, continuously monitors the ground conditions in real time and predicts the likelihood of ground subsidence to prevent accidents. Unlike typical buried electrode methods, the proposed system features a foldable electrode mechanism and a water supply device, thereby making installation and removal more efficient. Furthermore, it uses a ground stability determination algorithm that qualitatively assesses soft ground conditions, which are the primary cause of ground subsidence. The performance of the ERMS was validated through comparisons with commercial equipment, and its applicability was further confirmed through field tests conducted at mobile crane installations. The ERMS is expected to significantly reduce the risk of accidents caused by ground subsidence during mobile crane operations and to contribute to enhancing overall safety in construction environments.

This work applies geometric deep learning, a branch of artificial intelligence (AI), to the problem of understanding and structuring the latent representations used by image generative models. We propose a Hyperbolic Latent AutoEncoder (HLAE) that embeds image representations onto a product manifold (PM) of hyperbolic spaces . Unlike Euclidean latent spaces, the resulting PM latent space exhibits a distinctive geometric structure: semantic information concentrates in the directional component, while the radial component remains class-invariant. We quantify this structure through linear probing (41.3% vs. 10.2% for a Euclidean autoencoder (AE) baseline), direction-radius decomposition, and perturbation analysis. Through systematic generative exploration, centroid sampling, Gaussian sampling, and cross-class interpolation, we demonstrate that conventional strategies fail in PM latent spaces due to the manifold's curved structure. Notably, flow-based sampling produces class-diverse latent representations (entropy 2.22/2.30, classifier confidence 0.90) that remain within the learned PM latent distribution. Our analysis reveals the geometric constraints that PM latent spaces impose on generation, and provides empirical grounding for future geometry-aware generative models.

The construction industry requires data-efficient vision systems for autonomous machinery, yet labeled real-world data remains scarce. This study systematically investigates synthetic and real data composition for construction-specific object detection. A 3,200-image synthetic dataset is generated using the FLUX.2 Klein 9B model across four object types—steel pipe piles, concrete piles, H-beams, and timber piles. A factorial experimental design combines three synthetic scales (500, 2,000, and 3,200 images) with four real-data levels (0, 5, 10, and 30 images), yielding 13 trained YOLOv8n models evaluated on 90 real-world images. Results show that synthetic-only models fail to generalize despite high validation performance, while hybrid training significantly improves real-world detection. A tipping point is identified: as few as 5–10 real images per category achieve reliable performance. Diminishing returns in synthetic scaling and domain gap effects are quantified using CLIP-based embedding analysis. The findings provide actionable guidelines for data-efficient object detection in data-scarce construction environments.

Generative flow quality need not improve monotonically as numerical integration is refined when the learned velocity field is itself approximate. We investigate this behavior in Product Manifold Flow Matching (PMFM), which represents image patches on Products of hyperbolic factors and samples with exponential-map Euler updates. Across 32 M and 129 M CIFAR-100 models, Product at NFE = 4 achieves lower FID than the original Euclidean baseline at NFE = 32 in all six architecture–training-seed configurations, with an approximately eightfold reduction in measured sampling time. A targeted 32 M ablation shows, however, that this crossover is not specific to Product geometry: a factor-wise Euclidean analog of the Product target-speed reweighting improves the Euclidean baseline across the evaluated NFE range and reproduces the cross-budget advantage in all three training seeds. Removing the same reweighting from the Product target degrades FID from NFE ≥ 4. Product FID reaches an early optimum at NFE = 8 on CIFAR-100 and at NFE = 4 on a balanced synthetic pile-driver corpus, while same-initial-condition step-doubling discrepancies continue to decrease. These results separate numerical refinement from generative fidelity and identify target-speed reweighting as a major contributor to low-NFE quality, without supporting a curvature-only causal interpretation.

Generative synthetic imagery can broaden visual coverage while still providing invalid or incomplete supervision to an object detector. This study revisits a failed synthetic-data pipeline for pile detection and tests whether severe localization-label collapse was sufficient to explain its synthetic-to-real transfer failure. A provenance audit identified a clean corpus of 3200 FLUX.2 [klein] 9B-generated images whose detector labels all used the same normalized box (cx = 0.500, cy = 0.500, w = 0.315, h = 0.525). A balanced 320-image subset was frozen before manual correction and reannotated under a visual-identifiability criterion, yielding 1154 pile instances. The fixed template matched any corrected pile at IoU ≥ 0.50 in only 40/320 images (12.5%) and, by supplying one box per image, necessarily omitted at least 834/1154 corrected instances (72.3%). Four matched YOLOv8n supervision regimes were then compared across three seeds: fixed-template (F0), geometry-matched but image-permuted (P1), corrected-single (C1), and corrected-complete (CAll). C1 and CAll fitted their own synthetic supervision almost perfectly (mAP50 = 0.9949 and 0.9890, respectively), yet real-domain transfer did not recover: on test56, CAll achieved mAP50 = 0.0008 versus 0.0492 for F0; on the secondary 52-image set, CAll achieved 0.0019 versus 0.0237. A small-real-data reference trained on only 19 real images (66 instances) achieved 0.0563 and 0.0298 mAP50 on the same two real sets. A Faster R-CNN control reproduced the absence of corrected-label recovery on the primary real set while fitting CAll in-domain (AP50 = 0.9737). Thus, localization-label collapse was genuine and severe, but correcting it was not sufficient to restore transfer. Annotation integrity is necessary but not sufficient; residual synthetic-to-real mismatch must be evaluated independently of label geometry.