70,000 point cloud samples representing handwritten digits across 10 distinct classes. The data is partitioned into 60,000 training and 10,000 test instances derived from 28x28 grayscale images.
Use Cases
- Train 3D neural networks to classify digits using the 10-class labels
- Benchmark geometric deep learning models using the 60,000 training and 10,000 test samples
- Evaluate point cloud feature extraction techniques on the 28x28 spatial representations
Strengths
- 70,000 total samples with a uniform distribution of 7,000 points per class
- Standardized split of 60,000 training and 10,000 test points
- Data derived from 28x28 black-and-white source images