# Ridiculously Fast Shot Boundary Detection with Fully Convolutional Neural Networks

## TL;DR

This paper introduces a fully convolutional neural network for shot boundary detection that learns end-to-end from raw pixels, achieving state-of-the-art accuracy and over 120x real-time speed on large video datasets.

## Contribution

The authors propose a novel fully convolutional neural network architecture for shot boundary detection that processes large temporal contexts efficiently and is trained on a large automatically generated dataset.

## Key findings

- Achieves state-of-the-art shot boundary detection accuracy.
- Runs at more than 120 times real-time speed.
- Effectively leverages large temporal context without repeated processing.

## Abstract

Shot boundary detection (SBD) is an important component of many video analysis tasks, such as action recognition, video indexing, summarization and editing. Previous work typically used a combination of low-level features like color histograms, in conjunction with simple models such as SVMs. Instead, we propose to learn shot detection end-to-end, from pixels to final shot boundaries. For training such a model, we rely on our insight that all shot boundaries are generated. Thus, we create a dataset with one million frames and automatically generated transitions such as cuts, dissolves and fades. In order to efficiently analyze hours of videos, we propose a Convolutional Neural Network (CNN) which is fully convolutional in time, thus allowing to use a large temporal context without the need to repeatedly processing frames. With this architecture our method obtains state-of-the-art results while running at an unprecedented speed of more than 120x real-time.