Repository navigation
Download the dataset
das-perimeter-v1.0.2.zip (about 1.55 GB) contains all four aligned representations of 4,000 samples across five event classes:
- Five class-specific 2D MAT files and one combined 1D MAT file.
- STFT arrays and 15-dimensional feature vectors.
- Sample metadata, class labels and fixed train/validation/test splits.
- A root
README.txtand the full CC BY-NC 4.0LICENSEfor standalone use.
Check the archive against SHA256SUMS below. For standalone data use, extract it into a separate folder.
Get started
- Train an SVM without the full download: Open in Colab. The feature CSV and fixed splits are included in the repository; no GPU is needed.
- Get the current code and checkpoints: visit the repository or download its code ZIP.
- Use MATLAB or run the CNNs: follow the quick-start guide. The official downloader verifies and installs the data automatically.
The dataset ZIP does not contain code or checkpoints. Keep all four samples from the same recording in one subset and use the supplied splits for comparable benchmark results.
License and contact
- Code: MIT.
- Dataset, Model Checkpoints, and Documentation: CC BY-NC 4.0, including accompanying results and figures.
The MIT code license does not grant commercial rights to the data or trained weights. Commercial licensing inquiries: ying-huan.li@connect.polyu.hk.