Video Summarization Tool

Description

This tool generates video summaries using four state-of-the-art summarization models:

• PGL-SUM
• CA-SUM
• DSNet anchor based
• DSNet anchor free

The models are pretrained on the TVSum and SumMe datasets.

Installation

1. Set up a virtual environment:

   python -m venv .summarization

2. Activate the virtual environment:

   source .summarization/bin/activate

3. Install required packages:

   pip install -r requirements.txt

Usage

Summary Generation

1. Navigate to the source folder:

   cd src

2. Generate summary for a single video:

   python inference.py pglsum --source ../custom_data/videos/source_video_name.mp4 --save-path ./output/summary_video_name.mp4 --sample-rate 30 --final-frame-length 30

3. Generate summaries for a folder of videos:

   python inference.py pglsum --source ../custom_data/videos/source_video_folder --save-path ./output/summary_videos_folder --sample-rate 30 --final-frame-length 30

Model Name References

• pglsum - PGL-SUM
• casum - CA-SUM
• dsnet_ab - DSNet anchor based
• dsnet_af - DSNet anchor free

Parameters Explanation

• --sample-rate 30: The model analyzes every 30th frame.
• --final-frame-length 30: The resulting video summary will contain around 30 frames, roughly equivalent to 27 seconds. This duration can vary between 23-31 seconds depending on the frames per second of the original video.
• --max-shot-length 8: A single shot in the summary won't exceed 8 frames.
• --min-penalty-shot-length 5: Shots that are 5 frames or shorter will incur a length penalty, thus making them less likely to appear in the final summary.