CLIP-Sculptor: Zero-Shot Generation of High-Fidelity and Diverse Shapes from Natural Language

§ "a baseball cap" "a skateboard" "a motor bike" "a formula one car" "a truck that looks like a limo" "a travel bag" "an office chair" "an airplane" "a round shaped lamp" "a delta wing" Figure 1. CLIP-Sculptor is a zero-shot text-to-shape generation method. Left Top/Bottom: It generates diverse shapes that reflect the semantic meaning of the text input without requiring any text data during training. Right: The method also generates high-fidelity shapes.