Dataset Card for SportsShot

SportsShot is a dataset for shot segmentation and shot boundary detection.

Dataset Details

Dataset Description

Shot segmentation is an important and challenging task in video understanding. The existing benchmarks either simply focus on shot boundary detection or lack well-defined shot categories for segmentation. SportsShot is a fine-grained dataset for shot segmentation as well as shot boundary detection in multiple sports scenes.

• License: This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

Uses

SportsShot can be used for training shot segmentation / shot boundary detection models to automatically cut videos into meaningful clips.

Dataset Structure

The dataset contains 1200 videos and gt files for shot segmentation and shot boundary detection.

Gt files for shot segmentation contain the labels for all frames in all videos.

Gt files for shot boundary detection contain the starting frames and ending frames for all shot transitions.

Dataset Creation

Curation Rationale

The existing benchmarks either simply focus on shot boundary detection or lack well-defined shot categories for segmentation.

Source Data

Our SportsShot dataset focuses on sports scenes, so we search for volleyball, basketball, and football competition videos of levels like Olympics and World Cup on YouTube. These videos are recorded by professional high-definition cameras, guaranteeing excellent video quality. To ensure data balance across different sports, we manually select and cut these official competition records into 400 videos per sport using FFmpeg. Each video clip is in high definition, with a minimum resolution of 720p, and a consistent frame rate of 25fps.

Annotations

Annotation process

For data annotation, We recruit a team of 50 annotators and design an annotation tool. Our annotation tool provides a user-friendly interface that facilitates efficient and accurate data annotation. Videos loaded into the tool can be played at several different speeds, and with a double-click at the frame numbers stored as breakpoints or annotations, the annotators can jump to the exact frames instantly. The annotators can easily add, delete, sort, or edit their annotations. Moreover, the tool provides a quick check of the annotations, helping the annotators to find and correct some simple mistakes, such as overlaps between two segments and frames without any annotation.

Who are the annotators?

The annotators are college students.