DiverXplorer: Stock Image Exploration via Diversity Adjustment for Graphic Design

TL;DR

An image exploration prototype that enables stepwise adjustment of diversity, allowing users to transition from diverse overviews to increasingly focused subsets during exploration, and aims to provide a novel perspective on how to implement transitions between diversity and similarity.

Abstract

Graphic designers explore large stock image collections during open-ended or early-stage design tasks, yet common tools emphasize relevance and similarity, limiting designers' ability to overview the design space or discover visual patterns. We present an image exploration prototype that enables stepwise adjustment of diversity, allowing users to transition from diverse overviews to increasingly focused subsets during exploration. Our approach implements diversity control via determinantal point process (DPP)-based sampling and exposes diversity-similarity tradeoffs through interaction rather than static ranking. We report findings from a pilot study with professional graphic designers comparing our technique to baselines inspired by current tools in open-ended image selection tasks. Results suggest that stepwise diversity control supports early-stage sensemaking and comparison of visual patterns, while revealing important tradeoffs: diversity aids discovery and reduces backtracking, but becomes less desirable as exploration progresses. We aim to provide a novel perspective on how to implement transitions between diversity and similarity. Our code is available at https://github.com/CyberAgentAILab/DiverXplorer.

Figures (6)

• Figure 1: Scroll displays all images on a single screen, sorted by relevance to the task (i.e., "A design to celebrate Mother's Day aimed at young mothers with the feeling of "spending happy times with their children").
• Figure 2: Scroll+div displays all images on a single screen, and uses the DPP algorithm to rerank them by diversity and relevance to the task (i.e., "A design to celebrate Mother's Day aimed at young mothers with the feeling of "spending happy times with their children").
• Figure 3: Clustering generates a set of subclusters/steps (e.g., 3 in the image) for which only similar images are linked to. The blue arrow represents the selected image on each step. The hierarchy is data-dependent and a constant number of subclusters cannot be guaranteed (e.g., only four options in step 2).
• Figure 4: Ranking results. DiverXplorer was ranked 1st by more than 60% of the designers.
• Figure 5: The determinant maximization of DPP is equivalent to the maximization of the volume comprised by the vectors in the feature space. (a) The further from each other samples are in the feature space, the larger the vector area. (b) Case of more than two dimensions.