An implementation of the paper Skeleton-based gait index estimation with LSTMs
- Python
- Numpy
- TensorFlow
- Scikit-learn
- The code was implemented to directly work on DIRO gait dataset
- Please download the skeleton data and put the npz file into the folder dataset
python3 LSTM_AE_gait.py
The LSTM_AE class was modified from iwyoo's work
Default training and test sets
Finish loading data
(1000, 12, 17)
(400, 12, 17)
(3200, 12, 17)
Tensor("Placeholder:0", shape=(50, 12, 17), dtype=float32)
Training axis X with 100 epochs...
(50, 12, 17)
(50, 12, 17)
X-axis
AUC = 0.834
Training axis Y with 100 epochs...
(50, 12, 17)
(50, 12, 17)
Y-axis
AUC = 0.822
Training axis Z with 100 epochs...
(50, 12, 17)
(50, 12, 17)
Z-axis
AUC = 0.742
=== summation ===
per-segment (non-weighted sum)
AUC = 0.860
per-sequence (non-weighted sum)
AUC = 0.922
per-segment (weighted sum)
AUC = 0.904
per-sequence (weighted sum)
AUC = 0.953