Advertising Images Dataset for Visual Analysis

Abstract

This is a dataset of 606 ad images, manually gathered from different social media platforms, mainly from Egyptian digital ads. It can be used for research and application in visual analysis, marketing evaluation and machine learning. The data includes a variety of real-life advertisement styles in the Egyptian marketplace, with different combinations of colors, composition, typography, contrast and call-to-action (CTA) strategies frequently used in local digital marketing campaigns. We used a semi-automated approach for structured labeling. The process was assisted by a prompt, developed using marketing evaluation techniques. The prompt outlines scoring rules and label categories, which allow for systematic scoring of the images. Each image is labeled with the following criteria: score (0–10): overall quality based on color harmony, contrast, text clarity, layout balance, and CTA effectiveness color: qualitative assessment (e.g., good harmony, too many colors, consistent palette) contrast: visibility and readability evaluation text: extracted textual content from the advertisement layout: structural organization (e.g., balanced, cluttered, good hierarchy) cta: effectiveness of the call-to-action (e.g., strong CTA, unclear CTA, missing CTA) The prompt uses a fixed CSV format and predefined labels to keep the data clear and consistent. This method integrates human interpretation with specific guidance, leading to consistent and real-world relevant annotations. This dataset will be beneficial for students and researchers working in the fields of artificial intelligence, computer vision and digital marketing, particularly those focused on Egyptian advertising. It's suitable for regression, classification, and quality assessment of advertisements.