Uncharted Asian Facial Expressions: 52 MediaPipe Blendshapes for 7 Emotions

52-dimensional MediaPipe blendshape features extracted from web-crawled images of Asian facial expressions. The dataset is structured for seven distinct emotion categories. Its origin and specific collection details are not provided in the available metadata.

Use Cases

• Train facial expression recognition models based on 52-dimensional blendshape features.
• Benchmark emotion classification performance on Asian facial expression data.
• Analyze the relationship between specific MediaPipe blendshapes and the seven labeled emotions.
• Develop data augmentation or synthesis techniques for facial expression datasets.

Strengths

• Features are derived from a standardized computer vision pipeline (MediaPipe), suggesting consistent extraction.
• Data focuses on a specific demographic (Asian faces), which may address a gap in facial expression datasets.
• The description specifies a structured output of 52 features for seven emotion classes.

Limitations

• Row count and dataset scale are unknown, limiting suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data may reflect geographic or source bias inherent to web-crawled images from Kaggle.

Provenance

Source

kaggle

Collection Method

Features were extracted from web-crawled images using MediaPipe.

Geography

Likely contains Asian facial expressions, but specific countries or regions are not specified.