FindingEmo is a new image dataset built by KU Leuven and other institutions specifically for emotion recognition tasks.

The dataset contains annotations for 25,000 images. Unlike traditional datasets, FindingEmo focuses on complex situations involving multiple characters in natural scenes and annotates the emotional content of the entire scene, not just the emotions of individuals. The annotation dimensions include emotional value, arousal, and emotional labels, and these annotations are collected through the Prolific platform.

In addition, the dataset also provides URL links to the original images and all related source code. The FindingEmo dataset is a complex and challenging dataset that covers a rich range of emotional dimensions. It provides a new source of data for emotion recognition research and helps improve the generalization ability of the model and its performance in real scenarios.