Zedge Launches Comprehensive AI Training Dataset with 7,800 High-Quality Images

Zedge has launched DataSeeds.AISample Dataset, a comprehensive image collection for AI training, in collaboration with Perle.ai and Emet Research. The dataset comprises over 7,800 high-quality images, sourced from Zedge's photography game GuruShots, with detailed descriptions from expert reviewers and game players. This release marks a significant step towards generating accurate, human-reviewed data essential for modern AI advancements.

Zedge, Inc. (NYSE American: ZDGE) has announced the release of DataSeeds.AI Sample Dataset (DSD), a comprehensive image collection designed for AI training, in collaboration with Perle.ai and Émet Research. The dataset, comprising over 7,800 high-quality images sourced from Zedge's photography game GuruShots, includes detailed descriptions from expert reviewers and game players. This initiative marks a significant step towards generating accurate, human-reviewed data essential for modern AI advancements.

The DSD features 7,843 high-resolution, rights-cleared photographs, each selected through a unique in-game peer-ranking system that reflects aesthetic and compositional quality validated by a global photography community. Each image was enhanced with multi-tiered human annotation through Perle.ai's expert-in-the-loop pipeline, including pixel-level segmentation, structured scene descriptions, and technical metadata.

Key findings from the accompanying research paper indicate that AI models trained with DSD achieved 70% better results compared to benchmark datasets. Specifically, LLAVA-NEXT showed a 24.09% increase in BLEU-4 scores, demonstrating the dataset's effectiveness in improving model performance. When benchmarked against DSD annotations, AWS Rekognition achieved only a 0.19 F1 score, highlighting the limitations of automated commercial tagging systems.

Zedge's strategic entry into the AI training data market leverages its existing content ecosystem, positioning DataSeeds.AI as a major supplier for enterprises creating foundational AI models. The scalable content pipeline, with access to over 30 million rights-cleared images, enables Zedge to respond rapidly to enterprise AI training needs. This creates a competitive advantage in the data-centric AI development paradigm where quality training data is increasingly recognized as the differentiating factor in model performance.

For Zedge, this represents a significant business model expansion beyond consumer-facing digital content. By transforming its existing user-generated content (UGC) assets into enterprise-grade AI training resources, Zedge has created a new revenue stream that leverages its core community assets while diversifying beyond its traditional marketplace model.

The DSD and DataSeeds.AI platform offer a differentiated supplier of high-fidelity, human-reviewed datasets tailored to the evolving needs of the generative AI ecosystem. All data, models, and benchmarking results are reproducible and available on HuggingFace, encouraging adoption, validation, and further innovation.

References:
[1] https://www.stocktitan.net/news/ZDGE/zedge-s-data-seeds-ai-releases-foundational-dataset-for-computer-sckaxn99mqok.html
[2] https://www.nasdaq.com/press-release/zedges-dataseedsai-releases-foundational-dataset-computer-vision-and-generative-ai