An implementation of the paper Applying Adversarial Auto-encoder for Estimating Human Walking Gait Index
- Python
- Numpy
- TensorFlow
- Scikit-learn
- Matplotlib
- The code was implemented to directly work on DIRO gait dataset
- Please download the histogram data and put the npz file into the folder dataset
Process default training and test sets with suggested parameters
$ python3 main.py
Specify test subject for leave-one-out cross-validation, store sampled histograms and save AUC results
$ python3 main.py -l 0 -s 1 -f results.csv
| Param | Description |
|---|---|
| -l | index of test subject (0 to 8 for 9 subjects in DIRO gait dataset) |
| -e1 | first epoch for evaluation |
| -e2 | last epoch for evaluation |
| -o | overlapping segments (i.e. using sliding window) |
| -s | sampling histogram (GAN) and save images |
| -f | file for saving AUC results |
Saved file structure
| Test subject | Segment length | AUCs of partial and combined measures |
|---|---|---|
| 0 | 1 | AUC values |
| 0 | 10 | AUC values |
| ... | ..... | ..... |
| 0 | 1200 | AUC values |
Default training and test sets
(512, 1)
(?, 1)
training subjects: [0 2 4 5 8]
data shape:
(6000, 256)
(4800, 256)
(38400, 256)
Epoch 1: D_loss 9.980, G_loss 0.032, Recon_loss: 0.641
Epoch 2: D_loss 7.510, G_loss 0.025, Recon_loss: 0.499
.....
Epoch 309: D_loss 1.545, G_loss 0.616, Recon_loss: 0.328
Epoch 310: D_loss 1.550, G_loss 0.612, Recon_loss: 0.327
FINAL RESULTS (AVERAGE)
Results probability
( 1) AUC = 0.5952 (+0.0204)
( 10) AUC = 0.6319 (+0.0310)
.....
(1200) AUC = 0.6766 (+0.0404)
Results discriminator
( 1) AUC = 0.4238 (+0.0468)
( 10) AUC = 0.3965 (+0.0591)
.....
(1200) AUC = 0.3516 (+0.0872)
Results reconstruction
( 1) AUC = 0.8151 (+0.0074)
( 10) AUC = 0.8732 (+0.0079)
.....
(1200) AUC = 0.9524 (+0.0092)
Results dist + prob
( 1) AUC = 0.8142 (+0.0074)
( 10) AUC = 0.8724 (+0.0078)
.....
(1200) AUC = 0.9513 (+0.0085)
Results dist + disc
( 1) AUC = 0.8085 (+0.0104)
( 10) AUC = 0.8700 (+0.0111)
.....
(1200) AUC = 0.9594 (+0.0113)
Results combination
( 1) AUC = 0.8073 (+0.0103)
( 10) AUC = 0.8687 (+0.0110)
.....
(1200) AUC = 0.9583 (+0.0112)
Portions of the work employed codes from Agustinus Kristiadi and kevinroth.