retread

Detect reused frames in video with Python.

Retread works by computing a hash for each frame in a video (dhash, described here if you're interested). It then counts how many times in the video that frame is reused. Once the duplicates are calculated, the data is output in JSON format, suitable for use in e.g. D3.js.

A video with low reuse across the board doesn't repeat many frames - not many duplicated shots, and not very long static images either. For example, Mad Max Fury Road, a fast-moving film with short shots and quick cuts:

A video with high spikes reuses a lot of short shots or single images. For example, Memento, a film notable for its nonlinear story, cutting forward and backward across the same scenes:

Installation

Clone the repository:

$ git clone [email protected]:ryanfox/retread.git

Install dependencies:

$ cd retread
$ pip install -r requirements.txt

Usage

Quickstart

$ python retread.py my_video.mp4 > scores.json

Open bar.html in a browser to see the data plotted.

Advanced usage

Process a video file:

$ python retread.py my_video.mp4 > scores.json

scores.json will contain a list of integers, where each element is the number of duplicates that frame had in the original.

Processing a video can take a while - several minutes for a feature-length file. There are a lot of frames to analyze! If you want to reprocess the same video without waiting, you can use the flag -s (for "save") to save the hashes for later:

$ python retread.py -s my_video.mp4 > scores.json

The hashes will be saved in my_video.mp4.json. retread will overwrite the file if it already exists.

To reuse saved hashes, use the flag -u (for "use"):

$ python retread.py -u my_video.mp4.json > scores.json

This will read the saved hashes (much faster than re-processing the whole video), do the duplicate analysis, and write the duplicate counts to standard out as usual.

To see what the most-duplicated frames are, you can save a jpeg image of the N most common frames. Use the flag -c (for "common"):

$ python retread.py -c 5 my_video.mp4 > scores.json
$ ls
0.jpg
1.jpg
2.jpg
3.jpg
4.jpg
my_video.mp4
[etc...]

This can be combined with -u, but you still need to specify the video file):

$ python retread.py -u my_video.mp4.json -c 5 my_video.mp4 > scores.json
$ ls
0.jpg
1.jpg
2.jpg
3.jpg
4.jpg
my_video.mp4
my_video.mp4.json
[etc...]