Dog Emotion Recognition Dataset

This dataset focuses on recognizing emotions in dogs from images. While humans may find it easy to interpret animal emotions, accurately identifying them from photos is more complex, as dogs’ expressions can be subtle and not immediately obvious.

The dataset is based on five primary dog emotions: happy, sad, curious, relaxed/sleepy, and angry. Most of the images come from the Stanford Dogs Dataset, with a small portion collected additionally. The original dataset contained around 11,000 images covering 120 dog breeds. Due to class imbalance, data augmentation techniques were applied, and the dataset was expanded to approximately 30,000 images for training.

All images were manually annotated, since the original dataset did not include emotion labels. The labeling followed established guidelines describing how dogs express emotions through physical cues such as facial expressions, posture, and eye behavior. For example, happiness may be indicated by an open mouth and visible tongue, curiosity by a tilted head, anger by bared teeth and tense facial features, relaxation by closed eyes or yawning, and sadness by lowered head and flattened ears.

Overall, the dataset is a curated and augmented collection of dog images labeled into five emotional categories, designed for training models to recognize canine emotions from visual data.