SketchXplain: Intuitive Visual Explanations of Image Classifiers with Sketches

A team of researchers has proposed SketchXplain, a system that generates sketch-based visual explanations for image-based AI predictions, aiming to close an interpretability gap left by current methods. The work was submitted to the arXiv preprint server on 16 Jun 2026 [1]. The system addresses a core weakness of saliency maps, a common tool for explaining AI image classifiers. While saliency maps highlight influential regions in an image, the researchers note these visualizations are “often unintuitive and semantically unclear” [1]. To create more coherent explanations, SketchXplain combines techniques from saliency maps, concept-bottleneck models, and sketch optimization. The goal is to produce visual explanations that are intuitive, simple, and selective, inspired by the clarity of artistic drawings [2]. In evaluations, the system was tested on two distinct tasks: face expression recognition and skin lesion diagnosis [1]. User studies for the facial recognition task found that SketchXplain “supported quicker interpretation with more aligned visualizations than saliency maps or simple drawings” [2]. For skin lesion diagnosis, the sketch-based explanations more coherently visualized disease symptoms, which the study found better supported diagnosis by non-experts [2]. The paper was posted on arXiv, an open-access repository for electronic preprints that has been in operation since 1991 and now receives about 24,000 submissions per month [6]. The work appears under the Computer Science > Human-Computer Interaction category and is accessible through the arXivLabs framework, a platform that allows community collaborators to develop and share new features on the site [1][4]. arXivLabs, formalized in 2020, sets guidelines for collaborations to ensure partners share arXiv’s values of openness, community, excellence, and user data privacy [4]. The framework currently hosts a showcase of experimental tools, including bibliographic explorers and code-finding services, though arXiv announced a temporary pause on new proposals to focus on modernizing its core infrastructure [3][5].