Visualizes a graph stored as a json file as a force-directed graph in a webbrowser. It has specifically been built to visualize scene graphs, a representation commonly used in computer vision. See Visual Genome for details about the representation.

This type of visualization is often found in many computer vision papers that build models to display their model's output scene graphs.

First, you need an image and its associated scene graph.

Save a scene graph object as json in a file with the following format:

{
    url: STRING,
    objects: [
        {name: STRING},
        ...
    ],
    attributes: [
        {attribute: STRING, object: INT},
        ...
    ],
    relationships: [
        {predicate: STRING, object: INT, subject: INT},
        ...
    ]
}

The url should point to an image source. The attributes and relationships lists contain an object int field that is an index to the object it refers to in the objects list.

Run the following command to visualize the scene graph:

python visualize_scene_graph.py --graph LOCATION_OF_YOUR_SCENE_GRAPH

The above command will automatically open up the browser and display a scene graph and its associated image.

Feel free to contribute to this project by sending me a pull request. Feel free to follow me on twitter (@RanjayKrishna) for updates and to see my other work.