This repo provides code to perform video shot detection and extract middle, beginning and end frames for each shot.

You should only need to install PyAV library, which will install other required dependencies. conda install av -c conda-forge

The code has been tested om python 3.6

Compile using

cmake .
make

Extract the movie shots running

sh run_shot_detection.sh {input_movie_file} {output_folder}

The code will create a folder {output_folder}/{input_movie_file} containing the start, middle and last frame for every shot and a file {output_folder}/{input_movie_file}/{input_movie_file}.videvents with the frame and timestamp of every start of shot.

You can test the code using a test video as

wget http://wednesday.csail.mit.edu/frames/tools/video_tools/test_vid/short.mp4
mkdir output
sh run_shot_detection.sh short.mp4 output/

Extracts the movie .dfd file, capturing the difference between motion compensated frames.

./ShotDetection input_file.mp4 output_file.dfd

Creates mat_index containing frame-number and timestamp mappings and saves into some base dir

python mat_index.py --video_fname input_file.mp4 --base_dir out_dir

Create file with shot changes from .dfd and save into a base dir

python dfd_to_videoevents.py --base_dir out_dir --dfd_path output_file.dfd

Given the shot changes, dump the first middle and last frame for every shot.

python dump_frames_of_shot.py --video_fname input_file.mp4 --dfd_path output_file.dfd --base_dir "$out_dir"

If you want to enable parallel processing (of block matching) with OPENMP, please add -fopenmp to the CMAKE_CXX_FLAGS.

The code is based on CVHCI/Okapi library for processing of images.