Researchers have introduced CrowdH-Edit, a method for generating synthetic Hajj crowd images using reference-guided, instruction-based editing. The work responds to a persistent problem in computer vision: annotated images of Hajj crowds are scarce, making it difficult to develop reliable crowd-counting systems for this specific setting.
The abstract argues that existing crowd-counting datasets and unconstrained generative models are insufficient because they do not capture the distinctive conditions of Hajj crowds. By using reference images and textual instructions, CrowdH-Edit aims to produce more relevant synthetic training data.
The source is a short abstract, so it does not include quantitative results or comparisons with other methods. The contribution is framed as a data-generation strategy rather than a finished counting system.