README.md

---
task_categories:
- image-classification
license: mit
---

## Datasets
1. We conduct experiments on three new 3D domain generalization ([3DDG](#new-3ddg-benchmarks)) benchmarks proposed by us, as introduced in the next section. 
    - base-to-new class generalization (base2new)
    - cross-dataset generalization (xset)
    - few-shot generalization (fewshot)

2. The structure of these benchmarks should be organized as follows.
```sh
    /path/to/Point-PRC
    |----data # placed in the same level of `trainers`, `weights`, etc. 
        |----base2new
            |----modelnet40
            |----scanobjectnn
            |----shapenetcorev2
        |----xset
            |----corruption
            |----dg
            |----sim2real
            |----pointda
        |----fewshot
            |----modelnet40
            |----scanobjectnn
            |----shapenetcorev2
```

3. You can find the usage instructions and download links of these new 3DDG benchmarks in the following section. 

## New 3DDG Benchmarks
### _Base-to-new Class Generalization_
1. The datasets used in this benchmark can be downloaded according to the following links.
    - [ModelNet40](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/base2new/modelnet40)
    - [S-OBJ_ONLY](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/base2new/scanobjectnn/obj_only)
    - [S-OBJ_BG](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/base2new/scanobjectnn/obj_bg)
    - [S-PB_T50_RS](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/base2new/scanobjectnn/hardest)
    - [ShapeNetCoreV2](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/base2new/shapenetcorev2)

2. The following table shows the statistics of this benchmark.

    ![](assets/base-to-new.png)

### _Cross-dataset Generalization_
1. The datasets used in this benchmark can be downloaded according to the following links.
    - [OOD Generalization](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/xset/dg)
        - OmniObject3d (Omin3D)
    - [Data Corruption](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/xset/corruption)
        - ModelNet-C (7 types of corruptions)
            - add global outliers, add local outliers, dropout global structure, dropout local region, rotation, scaling, jittering
    - [Sim-to-Real](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/xset/sim2real)
    - [PointDA](https://huggingface.co/datasets/auniquesun/Point-PRC/tree/main/new-3ddg-benchmarks/xset/pointda)

2. The following table shows the statistics of this benchmark.

    ![](assets/cross-dataset.png)

### _Few-shot Generalization_
1. Although this benchmark contains same datasets as the _Base-to-new Class_, it investigates the model generalization under extremely low-data regime (1, 2, 4, 8, and 16 shots), which is quite different from the evaluation setting in _Base-to-new Class Generalization_.

2. The following table shows the statistics of this benchmark.

    ![](assets/few-shot.png)

Dataset presented in [Point-Cache: Test-time Dynamic and Hierarchical Cache for Robust and Generalizable Point Cloud Analysis](https://arxiv.org/abs/2503.12150). 

Code: https://github.com/auniquesun/Point-Cache