SofaOffscreenCamera plugin

Contact: Jean-Nicolas Brunet ([email protected])

This SOFA plugin brings an offscreen camera that is able to take screenshot with or without GUI. Multiple offscreen cameras can be added to a single simulation scene, and take screenshots from different angles at the same time.

Here is a quick example.

<OffscreenCamera 
    name="beam_and_ball"  
    filepath="camera_%s_%i.png" 
    save_frame_before_first_step="true" 
    save_frame_after_each_n_steps="5"
    multisampling="4"
    widthViewport="800"
    heightViewport="600"
    position="-20 0 0" lookAt="0 0 0" zNear="0.01" zFar="100" projectionType="1"/>

The multisampling data argument is an anti-aliasing parameter that will set the number of sampling points per pixel. Power of two should be used.

The filepath data argument will replace %s and %i for the component name and step number, respectively. In this example, a screenshot will be taken just before the simulation starts, and then at the end of every 5 steps. Hence, for a simulation of 10 steps, the following 800x600 images will be created:

1. camera_beam_and_ball_0.png
2. camera_beam_and_ball_5.png
3. camera_beam_and_ball_10.png

Warning: The option save_frame_before_first_step="true" will not work with a SOFA version v20.12 or less.

Offscreen camera should only render the component within their context tree. Hence, in the following example, the first camera will take a screenshot containing both the beam and the ball, while the second camera will only see the ball. In this example, both camera capture the same point of view, but this is not restricted.

<Node name="beam">
    (...)
    <HexahedronFEMForceField (...) />
    <OffscreenCamera
        name="beam_and_ball"
        filepath="camera_%s_%i.png"
        save_frame_before_first_step="true"
        position="-20 0 0" lookAt="0 0 0" zNear="0.01" zFar="100" projectionType="1"/>
    <Node name="ball">
        <OglModel (...) />
        <OffscreenCamera
            name="only_and_ball"
            filepath="camera_%s_%i.png"
            save_frame_before_first_step="true"
            position="-20 0 0" lookAt="0 0 0" zNear="0.01" zFar="100" projectionType="1"/>
    </Node>
</Node>

Scripting with python

Python bindings are also availabe. Hence, from a python script, you can manually call OffscreenCamera::save_frame(filepath). The position and lookAt data attributes can be dynamically changed at any time from the script to move around the camera. In the following example, a set of predefined positions is given to the camera at each time step. A screenshot is also rendered every time.

camera_positions = get_camera_positions()
camera = root.beam.camera
for i in range(n_frames):
        Sofa.Simulation.animate(root, 1)
        Sofa.Simulation.updateVisual(root)
        
        # Move the camera to the next position
        with camera.position.writeableArray() as wa:
            wa[0:3] = camera_positions[i+1]

        # Save screenshot at time i+1
        camera.save_frame(f'frame_{i+1}.jpg')

Note that this example could be rendered into an mp4 video using, for example, ffmpeg:

ffmpeg -framerate 60 -i 'frame_%d.jpg' -c:v libx264 -vf fps=30 -pix_fmt yuv420p simulation.mp4

Here, the input frame per second is set to 60, and the output to 30, which means that the video will last 2 times the simulation time.

The file examples/rotating_camera.py contains an example where two cameras are moved in an ellipse around the bending beam. The frames of both cameras are manually render into the "only_ball" and "beam_and_ball" directories, respectively. The python script also dynamically change the position of the camera after each simulation steps. The following result was computed by running the examples/rotating_camera.py file on an headless (no GUI available) server.