UVigo Video Classification Lab 02: A Subset of UCF101 for GPU Training

A small subset of the UCF101 action recognition dataset, designed for rapid 4D tensor manipulation and GPU VRAM efficiency. The dataset is hosted on Kaggle and is intended for educational or experimental use in computer vision. Its specific size, license, and authorship details are not provided.

Use Cases

• Prototyping video action recognition models based on the UCF101 subset.
• Testing 4D tensor manipulation pipelines for video data.
• Benchmarking GPU memory usage and training speed on video classification tasks.
• Educational demonstrations of video classification and action recognition concepts.

Strengths

• Derived from the established UCF101 benchmark for human action recognition.
• Designed for efficient GPU VRAM usage, which suggests optimization for training.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UCF101

Collection Method

Subsampled from the UCF101 dataset for specific lab purposes.