3D-JEPA: A Joint Embedding Predictive Architecture for 3D Self-Supervised Representation Learning

Benchmark Model Rank Results
3d-part-segmentation-on-shapenet-part3D-JEPAInstance Average IoU: 84.93Class Average IoU: 86.41
3d-point-cloud-classification-on-modelnet403D-JEPAOverall Accuracy: 94.0
3d-point-cloud-classification-on-scanobjectnn3D-JEPAOverall Accuracy: 89.52OBJ-BG (OA): 93.63OBJ-ONLY (OA): 94.49
few-shot-3d-point-cloud-classification-on-13D-JEPAOverall Accuracy: 97.6Standard Deviation: 2.0
few-shot-3d-point-cloud-classification-on-23D-JEPAOverall Accuracy: 98.8Standard Deviation: 0.4
few-shot-3d-point-cloud-classification-on-33D-JEPAOverall Accuracy: 94.3Standard Deviation: 3.6
few-shot-3d-point-cloud-classification-on-43D-JEPAOverall Accuracy: 96.3Standard Deviation: 2.4