Frames Calculation

Toatal Frames: 527599

Label	Count
0	6332
1	6919
2	228
3	11679
4	16131
5	756
6	94847
7	21017
8	171437
9	198253

NOTES

Dataset Visualization

• TSNE

Future Checklist:

Visulization

• Confusion Matrix
• Skeleton & Predicted label
• Accuracy curve

Model Candidates

• LSTM / GRU 25 frames * 2 sample rate
• TCN Non Causal largest I can fit, it is fully conventional
• TCN Causal
• Bi-Directional RNN all models take the most simple structure no drop out
• Fully Connected Layer

Evaluation Metrics

• Cross-Validation
• S5 Only

Even The Parameters

to make the parameter number the same

Find Out Where It Fails And Why and Give It A Solution

csv to HDF5 http://docs.h5py.org/en/latest/

refer: https://discuss.pytorch.org/t/how-to-speed-up-the-data-loader/13740/2

refer LMDB (Lightning Memory-mapped Database)