GenAI-MVS: Exercise Video Clips with Frame-Level Annotations

82 video clips of five exercise types, totaling 8,029 frames, annotated for temporal action classification and form analysis. The dataset, created by AvihaiNaam, is split into 54 training and 28 validation videos. It was last updated on Hugging Face in November 2025.

Use Cases

• Temporal action classification based on labeled exercise types (bench press, deadlift, dips, pullups, pushups)
• Exercise form analysis based on frame-level annotations
• Training and validating video-based action recognition models based on the provided train/validation split

Strengths

• Contains 82 videos with frame-level annotations, providing detailed temporal labels
• Includes 8,029 total frames across five distinct exercise classes
• Provides a predefined split of 54 training and 28 validation videos

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Dataset size in terms of rows and file formats is unknown, which may limit suitability assessment

Provenance

Source

huggingface

Freshness

Last updated 2025-11-10 08:52:14