Minimum Semantic Content (MSC) image dataset, a balanced dataset containing "ugly" images (alongside more traditionally appealing ones) of objects with little semantic content

The Minimum Semantic Content (MSC) dataset provides a structured resource aimed at isolating aesthetic judgments from cognitive (semantic) influences, making it particularly useful for advancing empirical studies on aesthetics. Unlike existing image datasets that predominantly feature beautiful images, the MSC dataset offers a balanced selection of 10,426 images, carefully curated span the whole range from "ugly" to "beautiful". The images were also selected to minimize semantic content. This reduction allows researchers to focus on perceptual factors underlying aesthetic judgments, avoiding the confounding influence of complex cognitive associations. Each image in the MSC dataset has been evaluated by 100 observers, ensuring robust data on aesthetic ratings. By including comparable amounts of "beautiful" and "ugly" images, the MSC dataset enables a re-examination of relationships between low-level image features and aesthetic valuations. This expanded aesthetic range offers insights that challenge or refine current models, demonstrating that limiting datasets to predominantly beautiful images can skew, exaggerate, or miss significant findings on the aesthetic response to image content.
Each of the 10,426 images in the MSC dataset has been evaluated by 100 observers, ensuring robust data on aesthetic ratings. By including a method for generating "ugly" images alongside more traditionally appealing ones, the MSC dataset enables a re-examination of relationships between low-level image features and aesthetic valuations.

We provide a .csv file with raw crowdsourcing results, with each column representing the number of votes received for each of the five aesthetic valuation categories (1 = very ugly, 5 = very beautiful). we also provide results from truncated histogram fittings to the crowdsourcing data. These columns include the final value of mu (mean), the final value of sigma (standard deviation), the final FVAL and ExitFlag values from Matlab’s fminsearch function, and the number of iterations required. Images were modified by observers. We provide categorization columns, specifying whether an image is “uglified,” “beautified,” “auto_uglified,” or “unmodified.” This section also indicates if an image served as the starting point for a modification (e.g., “uglified_original” denotes an image used as the basis for an uglification process). This structure ensures that users can easily identify the type and provenance of each image, as well as access comprehensive metadata and aesthetic ratings for further analysis.