We conduct experiments on three new 3D domain generalization (3DDG) benchmarks proposed by us, as introduced in the next section.
base-to-new class generalization (base2new)
cross-dataset generalization (xset)
few-shot generalization (fewshot)
The structure of these benchmarks should be organized as follows.
/path/to/Point-PRC
|----data # placed in the same level of `trainers`, `weights`, etc.
|----base2new
|----modelnet40… See the full description on the dataset page: https://huggingface.co/datasets/auniquesun/Point-PRC.
