karldd
commented
3 years ago Yes that is right, the face to face extrusion cannot exactly represent all shapes in the dataset. From 4.2 in the paper:
Of the design sequences in the reconstruction dataset, 59.2% can be directly converted to a face extrusion sequence. We estimate that approximately 80% of designs in our dataset can be reconstructed by finding alternative construction sequences.
Finding alternate representations that can represent all shapes is an interesting area for future exploration.
Hippogriff
In the paper, the proposed network works on faces, where two faces are selected and operation is predicted. It seems to me that shapes with subtraction operation where subtraction creates a cavity throughout the shape, cannot be represented by this strategy. An example posted below encircled in red.
Some of these subtraction cases can be represented using add operation using trimmed faces, but I think it is not always possible to represent all shapes in this dataset with this strategy.