Trpimir Jeronim Ježić

Understanding abstract concepts of aesthetics and semantics is crucial in developing reliable and verifiable systems for information retrieval or content generation. In this study we explore the space of use-cases where neural networks can aid in deepening the understanding of aesthetic and semantic theoretical principles, their relations, their practical relevance, and possibilities for automation of theory-aligned, and thus explainable, semantic or aesthetic evaluation. We start with pretrained object-recognition vision models, refitted and fine-tuned to discriminate between classes of fine art paintings along two core-semantic axes: abstract-concrete and iconic-symbolic. We then refit the models one more time using the CBM method to make models more reliable and their outputs interpretable in the frame of design- and art-theoretic principles. Finally, we apply a curated collection of xAI methods to analyze, visualize, and contrast the decision-making criteria of four developed models. Our findings show that - for the task of mapping sets of highly distributed, complex and intertwined features - the introduction of narrow concept bottlenecks at the output end of the black-box networks affords for interpretation of models’ outputs, while leaving no negative impact on their classification performance. The methodology developed for this paper presents a transferable framework for amplifying black-box vision-models’ usability and verifiability of their outputs, and it is especially applicable in specialized fields with well-formed theoretical understanding of their respective imaging problem-space. All code developed for constructing, refitting and analyzing models, alongside the datasets created for this study, have been made publicly available under the GPLv3 license.

Although modern web technologies increasingly rely on high-performance rendering methods to support rich visual content across a range of devices and operating systems, the field remains significantly under-researched. The performance of animated visual elements is affected by numerous factors, including browsers, operating systems, GPU acceleration, scripting load, and device limitations. This study systematically evaluates animation performance across multiple platforms using a unified set of circle-based animations implemented with eight web-compatible technologies, including HTML, CSS, SVG, JavaScript, Canvas, and WebGL. Animations were evaluated under controlled feature combinations involving random motion, distance, colour variation, blending, and transformations, with object counts ranging from 10 to 10,000. Measurements were conducted on desktop operating systems (Windows, macOS, Linux) and mobile platforms (iOS, Android), using CPU utilisation, GPU memory usage, and frame rate (FPS) as key metrics. Results show that DOM-based approaches maintain stable performance at 100 animated objects but exhibit notable degradation by 500 objects. Canvas-based rendering extends usability to higher object counts, while WebGL demonstrates the most stable performance at large scales (5000–10,000 objects). These findings provide concrete guidance for selecting appropriate animation technologies based on scene complexity and target platform.

The research field of computational aesthetics gives crucial contributions to the development of mechanisms for filtering and/or generating value-laden informational content. This paper acknowledges a recognized escalating problem in the development of contemporary informational technologies and presents a practical solution for communicational quality management by employing an innovative approach to the computational aesthetic evaluation (CAE). After discussing the problem and attempted approaches to its alleviation, the paper offers a novel expert solution by presenting an original research approach and its resulting open-sourced model which outperforms its current state-of-the-art competition in semantic and stylistic classification, at the same time providing an idiomatic measure for objective aesthetic evaluation and demonstrating semantically rich and professionally recognized explanatory power which can serve as the solid basis for development of reliable and user friendly content retrieval, generative or auxiliary design applications. Presented model is resource- and privacy-wise utmost conservative. Its use evades all ethical, legal or security concerns that beset all currently prominent models. Its developmental and operational costs are practically nil.