Scenescoop Save

A tool to describe the content of videos and suggest similar scenes in other videos/films.

Project README


Scenescoop is a tool to get similar semantic scenes from a pair of videos. Basically, you input a video and get a scene that has a similar meaning in another video. You can run it as a python script or as a web app.


How it works

Scenescoop uses the im2text tensorflow model to analyze videos on a frame to frames basis and get a description of the content of those images. Frames with the same description are grouped together to create a sequence or scene.

Scenes are then analyzed with spaCy, for sentence parsing and built-in word vectors, using the average of the word vectors in the sentence.

Annoy is finally used to create an index for fast nearest-neighbor lookup (based on @aparrish Plot to poem)

This project is inspired by Thingscoop.

Video Demos

A man sitting at a table with a plate of food

A man sitting at a table with a plate of food

A group of people walking down the street

A group of people walking down the street


To run this you'll need to install a few dependencies. You can follow the original repository or the instructions Edouard Fouché wrote. (I plan to write a step-by-step guide on how to install everything)

You can also get the pretrained model I'm using here.

Once everything is installed, clone the repo and install the project dependencies:

git clone
cd scenescoop
pip install -r requirements.txt

You can then run Scenescoop in two modes:

1) Frame Analysis Mode

Given a video file --video (.mp4, .avi, .mkv or .mov), this will analyse the file frame by frame and output a .json file containing the descriptions of the those frames. The --name argument should be the output name of the transcript.


python --video videos/moonrisekingdom.mp4 --name moonrisekingdom

The .json file should look something like this:

"a person is taking a picture of themselves in a mirror ": [4834], 
"a man sitting in the back of a pickup truck ": [2265, 2266], 
"a man sitting on a bench in front of a building ": [1935, 1937, 
1938, 3950, 3951, 3952, 3953, 3960, 4072, 4073, 4074, 4075, 
4077, 4079, 4080, 4082, 4115, 4467], 
"a man standing next to a tree holding a surfboard ": [2470]

2) Transfer Mode

Two videos are required for this mode and both should have their corresponding transcript.json file created in the Frame Analysis Mode.

The --input_data argument should be the .json file containing the data for the input video and --transform_data is the .json file for the transfer video. --input_seconds is the input time frame to transfer and --transform_src is the video source of the transfer video.


python --input_data transcripts/street.json --input_seconds 0,5 --transform_src videos/her.avi --transform_data transcripts/her.json

You can print all options with python -h:

usage: [-h] [--video VIDEO] [--name NAME]
                     [--input_data INPUT_DATA] [--input_seconds INPUT_SECONDS]
                     [--transform_src TRANSFORM_SRC]
                     [--transform_data TRANSFORM_DATA] [--api API]


optional arguments:
  -h, --help            show this help message and exit
  --video VIDEO         Video Source to transform
  --name NAME           Name of the video
  --input_data INPUT_DATA
                        Input Video. Must be a json file.
  --input_seconds INPUT_SECONDS
                        Input Video Seconds to create transformation. Example:
  --transform_src TRANSFORM_SRC
                        Transform Video Source.
  --transform_data TRANSFORM_DATA
                        Transform Video Data. Must be a json file.
  --api API             API Request

Web App

You can also launch an interactive web app, using a flask server, to run the Frame Analysis Mode and Transfer Mode in a webpage. You'll still need all the dependencies installed.


To run the app in a local server:


The visit localhost:8080.

To modify the source code:

cd static
yarn watch


Local development of the MMS application:

Start ngrok

./ngrok http 7676

Configure the url in Twilio and in the server in NGROK_URL

Start the Redis server


Start the Celery worker:

celery -A server.celery worker

Finally start the server




Open Source Agenda is not affiliated with "Scenescoop" Project. README Source: cvalenzuela/scenescoop
Open Issues
Last Commit
1 year ago

Open Source Agenda Badge

Open Source Agenda Rating