While vision-language models (VLMs) have advanced into detailed image description, evaluation remains a challenge. Standard metrics (e.g. CIDEr, SPICE) were designed for short texts and tuned to recognize errors that are now uncommon, such as object misidentification. In contrast, long texts require sensitivity to attribute and relation attachments and scores that localize errors to particular text spans. In this work, we introduce PoSh, a metric for detailed image description that uses scene graphs as structured rubrics to guide LLMs-as-a-Judge, producing aggregate scores grounded in fine-grained errors (e.g. mistakes in compositional understanding). PoSh is replicable, interpretable and a better proxy for human raters than existing metrics (including GPT4o-as-a-Judge). To validate PoSh, we introduce a new dataset, DOCENT. This novel benchmark contains artwork, paired with expert-written references, and model-generated descriptions, augmented with granular and coarse judgments of their quality from art history students. Thus, DOCENT enables evaluating both detailed image description metrics and detailed image description itself in a challenging new domain. We show that PoSh achieves stronger correlations (+0.05 Spearman ρ) with the human judgments in DOCENT than the best open-weight alternatives, is robust to image type (using CapArena, an existing dataset of web imagery) and is a capable reward function, outperforming standard supervised fine-tuning. Then, using PoSh, we characterize the performance of open and closed models in describing the paintings, sketches and statues in DOCENT and find that foundation models struggle to achieve full, error-free coverage of images with rich scene dynamics, establishing a demanding new task to gauge VLM progress. Through both PoSh and DOCENT, we hope to enable advances in important areas such as assistive text generation.
The Index of American Design, a 1930s–40s federal work relief project, generated over 18,000 watercolors and extensive documentation now held at the National Gallery of Art. Reconciling thousands of associated constituents in the museum’s Collection Management System (CMS) with Wikidata using OpenRefine exposed both opportunities and challenges: incomplete or inconsistent identifiers, conflicts between local records and Wikidata, and the labor of aligning diverse roles beyond artists. This discussion paper examines those methodological hurdles, highlights strategies for improving reconciliation, and demonstrates how a linked Power BI tool enables ongoing monitoring of data quality, access, and representation across humanities research infrastructures.
Chemical imaging of Johannes Vermeer’s Woman Holding a Balance and A Lady Writing has augmented understanding of Vermeer’s painting process. When combined with previous magnified examination of the paint surface and limited sample analysis, reflectance and X-ray fluorescence imaging spectroscopies allowed for visualizing the texture of Vermeer’s underpaint, an early stage that approximated the colors of the final image. The presence of brushmarked underpaint suggests Vermeer worked quickly during this initial painting stage—a shift from previous scholarly understanding, based on surface appearance, that Vermeer worked slowly and smoothly throughout his painting process. This paper also presents pigment maps for surface paint and underpaint in the flesh tones of the women’s faces, their jackets, and the tablecloths. The results address how Vermeer achieved subtle color effects by adding or layering specific pigments. Finally, this paper further characterizes the painted sketch and discusses the challenges in imaging this sub-surface design layer.
