FineHARD is a high-quality image-text alignment dataset that was open-sourced by the 360 AI research team in 2025. The related paper results are:FG-CLIP: Fine-Grained Visual and Textual Alignment".

This dataset is characterized by scale and refinement. It contains 12 million images and their corresponding long and short description texts, covering 40 million bounding boxes. Each bounding box is accompanied by a fine-grained area description, and innovatively introduces 10 million sets of hard fine-grained negative samples. These interference samples screened by the algorithm can effectively improve the model's ability to distinguish similar targets.