GEOPARD : Geo metric P retraining for Ar ticulation Pre d iction in 3D Shapes
Pradyumn Goyal Dmitry Petrov Sheldon Andrews Yizhak Ben-Shabat, Hsueh-Ti Derek Liu Evangelos Kalogerakis · 2025
Read GEOPARD : Geo metric P retraining for Ar ticulation Pre d iction in 3D Shapes by Pradyumn Goyal Dmitry Petrov Sheldon Andrews Yizhak Ben-Shabat, Hsueh-Ti Derek Liu Evangelos Kalogerakis (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-02747.txt.
Description
Pradyumn Goyal and co-authors teach a model to predict how 3D objects bend and move, in a 2025 paper. Many objects, from laptops to cabinets, have parts that swing or slide. Guessing those motions from a single still shape is hard, and hand-labeled training data is scarce.
Their system, called GEOPARD, first learns without labels. A geometry-driven search finds part motions that are physically valid, with no parts tearing off or crashing into each other. Those found motions serve as practice for a transformer model. On a standard test set of movable objects, the method reached top scores. The work shows that shape alone can teach plausible motion.
Catalog record
| CREATOR | Pradyumn Goyal Dmitry Petrov Sheldon Andrews Yizhak Ben-Shabat, Hsueh-Ti Derek Liu Evangelos Kalogerakis |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-02747