VideoSummarization

Pre-requisite

Have the dataset from here available in the dataset directory, and modify the src/qvpipe_dataset.py if needed.

Usage Details

Preprocess Data with Frame Extraction Methods

cd src
python3 frame_extraction.py

Train and Test

cd src
python3 main.py --epochs 50 --classes 17 --num_frames 5 --train --valid --early_stopping --frame_selection random

Options

--epochs: Number of epochs for training.

--classes: 17 is the only implementation available.

--num_frames: Number of frames to represent a video.

--train: Flag to enable training.

--valid: Flag to enable testing.

--early_stopping: Flag to enable early stopping.

--frame_selection: Standalone options can be ["random", "less_blur", "motion", "k_means", "histogram", "entropy"]. Combination can also be used like ["less_blur motion", "motion k_means entropy", etc.,].