---
dataset_info:
  features:
  - name: video_id
    dtype: string
  - name: caption
    dtype: string
  - name: start_time
    dtype: float64
  - name: end_time
    dtype: float64
  - name: duration
    dtype: float64
  splits:
  - name: train
    num_bytes: 1523961
    num_examples: 14778
  download_size: 860794
  dataset_size: 1523961
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# ActivityNet-Captions

This dataset repo contains a curated subset of the ActivityNet-Captions dataset.



\\(\text{Filtering Logic  }\begin{cases}
\text{Video length }\leq \text{ 10 seconds:} & \text{Keep} \\
\text{Video length } > \text{ 10 seconds:} & \text{Discard}
\end{cases}\\)

## Analysis


<img width="500" src="https://cdn-uploads.huggingface.co/production/uploads/66d78facde54fea8a009927e/NUHx5WkWTGYgHFSHma7mn.png">



There are **10,759** unique action-segment labels in the `activitynet.json` file.