Part Segmentation on Scan2CAD (test)
0.021CD (chair)Neural Part Priors
Evaluation Results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Neural Part PriorsOptimizer=Adam, Learning Rate=3e-4/3e-5, Iterations=6002022.03 | 0.021 | 0.051 | 0.076 | 0.094 | 0.141 | 0.025 | 0.068 | 0.038 | 48.9 | 54.2 | 27.2 | 38.6 | 47.6 | 34 | 41.6 | 46.1 | |
| MLCVNet + StructureNetDetector=MLCVNet, Part Priors=StructureNet2022.03 | 0.025 | 0.101 | 0.092 | 0.09 | 0.359 | 0.041 | 0.118 | 0.059 | 48 | 35.6 | 19.5 | 33.1 | 26.7 | 34.2 | 32.9 | 41.3 | |
| Bokhovkin et al.Method Type=Coarse Part Decomposition2022.03 | 0.027 | 0.059 | 0.095 | 0.102 | 0.207 | 0.031 | 0.087 | 0.049 | 54.8 | 54.2 | 22.4 | 32.8 | 44.2 | 37.5 | 40.9 | 49 | |
| SG-NN + MLCVNet + PointGroupDetector=MLCVNet, Segmentor=PointGroup, Geometric Association=SG-NN2022.03 | 0.056 | 0.121 | 0.11 | 0.161 | 0.406 | 0.034 | 0.148 | 0.088 | 25.7 | 39 | 30 | 22.9 | 30.6 | 48.8 | 32.8 | 29.3 |