Synthetic data · Blender
Training data you cannot film: synthetic railway obstacles in Blender
You cannot put a rock in front of a train to collect training data. The scene is generated in Blender instead, and the labels come out with it: rock, animal and box on the track, each with an automatically generated box and class label.
SYNTHETIC DATA
The footage was generated in Blender and is not a field recording. The boxes are ground truth produced at render time, not model predictions. Internal R&D work.
The film has no dialogue, only sound design.
01What the film shows
The bounding boxes are not a model's predictions. They are ground truth produced automatically during rendering: the position, scale and class of every object in the scene are already known, so the label file falls out of the render. No frame-by-frame annotation, no human in the loop.
Obstacle-on-track events are exactly the data an autonomous railway system needs and exactly the data nobody has: they are rare, dangerous and impossible to stage. Synthetic generation turns the class distribution into something you choose rather than something you inherit. This is R&D work, not a deployed product; the footage is generated, not recorded in the field.
