Datasets We evaluate on four datasets ranging from 2D objects (MNIST [11]), 3D objects (ModelNet40 [31] rigid object, SHREC15 [12] non-rigid object) to real 3D scenes (ScanNet [5]). Object classification is evaluated by accuracy. Semantic scene labeling is evaluated by average voxel classification accuracy following [5]. We list below the experiment setting for each dataset: • MNIST: Images of handwritten digits with 60k training and 10k testing samples. • with 9,843 shapes for training and 2,468 for testing.ModelNet40: CAD models of 40 categories (mostly man-made). We use the official split • SHREC15: 1200 shapes from 50 categories. Each category contains 24 shapes which aremostly organic ones with various poses such as horses, cats, etc. We use five fold cross validation to acquire classification accuracy on this dataset. • ScanNet: 1513 scanned and reconstructed indoor scenes. We follow the experiment settingin [5] and use 1201 scenes for training, 312 scenes for test. 对这个实验用中文概括一下

这个实验使用了四个数据集进行评估，涵盖了2D对象（MNIST）、3D对象（ModelNet40刚性对象、SHREC15非刚性对象）以及真实的3D场景（ScanNet）。对于对象分类任务，使用准确率作为评估指标；对于语义场景标记任务，使用平均体素分类准确率进行评估，遵循了ScanNet论文中的实验设置。以下是每个数据集的实验设置：

1. MNIST：手写数字图像数据集，包含60,000个训练样本和10,000个测试样本。
2. ModelNet40：40个类别的CAD模型数据集（主要是人工制品）。使用官方的数据集划分，训练集包含9,843个形状，测试集包含2,468个形状。
3. SHREC15：来自50个类别的1,200个形状数据集。每个类别包含24个形状，主要是各种有机物的形状，如马、猫等。使用五折交叉验证获取该数据集上的分类准确率。
4. ScanNet：共有1,513个扫描和重建的室内场景数据。按照[5]论文的实验设置，使用1,201个场景作为训练集，312个场景作为测试集。

这些实验数据集用于评估作者提出的方法在不同领域和场景下的性能。