We present VizGenie, a self-improving, agentic framework that advances scientific visualization through large language model (LLM) by orchestrating of a collection of domain-specific and dynamically generated modules. Users initially access core functionalities-such as threshold-based filtering, slice extraction, and statistical analysis-through pre-existing tools. For tasks beyond this baseline, VizGenie autonomously employs LLMs to generate new visualization scripts (e.g., VTK Python code), expanding its capabilities on-demand. Each generated script undergoes automated backend validation and is seamlessly integrated upon successful testing, continuously enhancing the system's adaptability and robustness. A distinctive feature of VizGenie is its intuitive natural language interface, allowing users to issue high-level feature-based queries (e.g., "visualize the skull" or "highlight tissue boundaries"). The system leverages image-based analysis and visual question answering (VQA) via fine-tuned vision models to interpret these queries precisely, bridging domain expertise and technical implementation. Additionally, users can interactively query generated visualizations through VQA, facilitating deeper exploration. Reliability and reproducibility are further strengthened by Retrieval-Augmented Generation (RAG), providing context-driven responses while maintaining comprehensive provenance records. Evaluations on complex volumetric datasets demonstrate significant reductions in cognitive overhead for iterative visualization tasks. By integrating curated domain-specific tools with LLM-driven flexibility, VizGenie not only accelerates insight generation but also establishes a sustainable, continuously evolving visualization practice. The resulting platform dynamically learns from user interactions, consistently enhancing support for feature-centric exploration and reproducible research in scientific visualization.

Abstract We formalize Schrödinger's definitions of hue, saturation, and lightness, building on the foundational idea from Helmholtz that these perceptual attributes can be derived solely from the perceptual metric. We identify three shortcomings in Schrödinger's approach and propose solutions to them. First, to encompass the Bezold‐Brücke effect, we replace the straight‐line definition of stimulus quality between a color and black with the geodesic path in perceptual color space. Second, to model diminishing returns in color perception, we employ a non‐Riemannian perceptual metric, which introduces a potential ambiguity in defining lightness, but our experiments show that this ambiguity is inconsequential. Third, we provide a geometric definition of the neutral axis as the closest color to black within each equal‐lightness surface—a definition feasible only in a non‐Riemannian framework. Collectively, our solutions provide the first comprehensive realization of Helmholtz's vision: formal geometric definitions of hue, saturation, and lightness derived entirely from the metric of perceptual similarity, without reliance on external constructs.

, model these aspects of color perception, colormaps are still mostly evaluated through piecewise linear interpolation in a Euclidean color space. In a non-Euclidean setting, the piecewise linear interpolation of a colormap through control points translates to finding shortest paths. Alternatively, a smooth interpolation can be generalized to finding the straightest path. Both approaches are difficult to solve and are compute intensive. We compare the 11 most promising optimization algorithms for the computation of a geodesic either as the shortest or as the straightest path to find the most efficient one to use for colormap interpolation in real-world applications. For two control points, the zero curvature algorithms excelled, especially the 2D relaxation method. For multiple control points, only the mimimal curvature algorithms can produce smooth curves, amongst which the 1D relaxation method performed best.

Scientists nowadays use data sets generated from large-scale scientific computational simulations to understand the intricate details of various physical phenomena. These simulations produce large volumes of data at a rapid pace, containing thousands of time steps so that the spatiotemporal dynamics of the modeled phenomenon and its associated features can be captured with sufficient detail. Storing all the time steps into disks to perform traditional offline analysis will soon become prohibitive as the gap between the data generation speed and disk I/O speed continues to increase. In situ analysis, i.e., in-place analysis of data when it is being produced, has emerged as a solution to this problem. In this work, we present an information-theoretic approach for in situ reduction of large-scale time-varying data sets via a combination of key and fused time steps. We show that this approach can greatly minimize the output data storage footprint while preserving the temporal evolution of data features. A detailed in situ application study is carried out to demonstrate the in situ viability of our technique for efficiently summarizing thousands of time steps generated from a large-scale real-life computational simulation code.

As the computation power of supercomputers increases, so does simulation size, which in turn produces orders-of-magnitude more data. Because generated data often exceed the simulation's disk quota, many simulations would stand to benefit from data-reduction techniques to reduce storage requirements. Such techniques include autoencoders, data compression algorithms, and sampling. Lossy compression techniques can significantly reduce data size, but such techniques come at the expense of losing information that could result in incorrect post hoc analysis results. To help scientists determine the best compression they can get while keeping their analyses accurate, we have developed Foresight, an analysis framework that enables users to evaluate how different data-reduction techniques will impact their analyses. We use particle data from a cosmology simulation, turbulence data from Direct Numerical Simulation, and asteroid impact data from xRage to demonstrate how Foresight can help scientists determine the best data-reduction technique for their simulations.

Abstract Applying lossy data compression to climate model output is an attractive means of reducing the enormous volumes of data generated by climate models. However, because lossy data compression does not exactly preserve the original data, its application to scientific data must be done judiciously. To this end, a collection of measures is being developed to evaluate various aspects of lossy compression quality on climate model output. Given the importance of data visualization to climate scientists interacting with model output, any suite of measures must include a means of assessing whether images generated from the compressed model data are noticeably different from images based on the original model data. Therefore, in this work we conduct a forced‐choice visual evaluation study with climate model data that surveyed more than one hundred participants with domain relevant expertise. In addition to the images created from unaltered climate model data, study images are generated from model data that is subjected to two different types of lossy compression approaches and multiple levels (amounts) of compression. Study participants indicate whether a visual difference can be seen, with respect to the reference image, due to lossy compression effects. We assess the relationship between the perceptual scores from the user study to a number of common (full reference) image quality assessment (IQA) measures, and use statistical models to suggest appropriate measures and thresholds for evaluating lossily compressed climate data. We find the structural similarity index (SSIM) to perform the best, and our findings indicate that the threshold required for climate model data is much higher than previous findings in the literature.

Pseudocoloring is one of the most common techniques used in scientific visualization. To apply pseudocoloring to a scalar field, the field value at each point is represented using one of a sequence of colors (called a colormap). One of the principles applied in generating colormaps is uniformity and previously the main method for determining uniformity has been the application of uniform color spaces. In this paper we present a new method for evaluating the feature detection threshold function across a colormap. The method is used in crowdsourced studies for the direct evaluation of nine colormaps for three feature sizes. The results are used to test the hypothesis that a uniform color space (CIELAB) will accurately model colormapped feature detection thresholds compared to a model where the chromaticity components have reduced weights. The hypothesis that feature detection can be predicted solely on the basis of luminance is also tested. The results reject both hypotheses and we demonstrate how reduced weights on the green-red and blue-yellow terms of the CIELAB color space creates a more accurate model when the task is the detection of smaller features in colormapped data. Both the method itself and modified CIELAB can be used in colormap design and evaluation.

A myriad of design rules for what constitutes a "good" colormap can be found in the literature. Some common rules include order, uniformity, and high discriminative power. However, the meaning of many of these terms is often ambiguous or open to interpretation. At times, different authors may use the same term to describe different concepts or the same rule is described by varying nomenclature. These ambiguities stand in the way of collaborative work, the design of experiments to assess the characteristics of colormaps, and automated colormap generation. In this paper, we review current and historical guidelines for colormap design. We propose a specified taxonomy and provide unambiguous mathematical definitions for the most common design rules.