Implement Classifier model with Auto Encoder.
$sudo docker build -t ae_classify .
$sudo docker run -it --gpus all ae_classify /bin/bash
# in the container below
# Train and evaluate
python3 main.py
# If want to save model weights, use save option
python3 main.py --save
# Only test with trained model
python3 test.py- Auto Encoder
Encoder -> Hidden Layer -> Decoder
- Fully Connected Layer
Hidden Layer(from Auto Encoder) -> NN -> 10 dims outputs
Implement 2 parts, AutoEncoder part and Classification part.
CIFAR-10 Dataset
Max Size 2500 images
2.Bird
4.Deer
9.Truck
Max Size 5000 images
- Others
1000 for each class.
| Algorithm | Mean | Std |
|---|---|---|
| Simple CAE | 31.6 | 10.818 |
| Over Sampled CAE | 72.8 | 0.4 |
| Under Sampled CAE | 69.6 | 0.489 |
| 2 Ensemble Under Sampled CAE | 75.0 | 0.632 |
| 3 Ensemble Over and Under Sampled CAE | 77.6 | 0.489 |
Pretrained model is available here