Pre_Data

Data preparation code to provide consistent and high-performance processing

dep

numpy tqdm numba joblib scipy

dataset

1. UAV-Human: Skeleton --> N C T V M
2. NTURGB-D：Skeleton [ST-GCN] --> N C T V M
3. NTURGB-D：Skeleton [CTR-GCN]
4. NW-UCLA：Skeleton --> N C T V M
5. MIMII：Audio --> [SNR[MFCC,device,label]]
6. SHL-2024: seqence --> Modal Channel Num sample
7. ECG5000: seqence
8. Kuairec
9. Tenrec

Target needs

1. Mem：Try to minimize memory consumption
2. Time：Faster as well as possible

Be care

1. Data preprocessing uses a lot of performance optimizations, the goal of which is trying to strike a reasonable balance between speed and demand
2. If you are willing to bear the memory consumption, you can replace 'open_memmap' with 'np.load' or 'np.save' to speed up the processing
3. This project has a high demand for I/O to ensure that it works on a medium with high I/O capability
4. Some datasets have different preprocessing patterns in different projects, and to avoid ambiguity, I have indicated the source in "[]".
5. N C T V M is Num Channel Frames Joint Body