We parse the input and output scenes into objects.
For all but one input object, there is only one object in the output scene with a matching position and shape, while one object is always deleted, making it straightforward to produce a mapping from input objects to output objects.
e.g.
- We can find rules by grouping objects by color.
- When we do this, we notice that:
- All light blue objects are replaced by another image.
- However, the replacement image varies by example.
- We notice that the replacement image is always the image of the non-light-blue object.
- All objects that aren’t light blue are deleted.
- All light blue objects are replaced by another image.
