This script extracts well-known Deep Learning features from DTD Dataset using the models and weights in keras. Please see the keras documentation for details on model architectures, data and training procedures used.
Currently, the script has support for Xception, VGG16, VGG19, ResNet50, InceptionV3, InceptionResNetV2, MobileNet, MobileNetV2, DenseNet121, DenseNet201, NASNetMobile, NASNetLarge and EfficientNetV2L and can be easily extended to extract further networks in keras.
Please use the requirements.txt file to install the required dependencies. It can be done with:
$ pip install -r requirements.txt
It is extremely recommended that you use a virtual environment to execute the line above.
To extract features using the VGG16 architecture:
$ python3 main.py -a VGG16
The resulting structure of the VGG16 folder will be:
VGG16/
-> Fold_1.txt
-> Fold_2.txt
-> Fold_3.txt
-> Fold_4.txt
-> Fold_5.txt
-> Fold_6.txt
-> Fold_7.txt
-> Fold_8.txt
-> Fold_9.txt
-> Fold_10.txt
-> weights.best.hdf5
where fold-X.txt contains the accuracy for fold X. The weights.best.hdf5 file, on the other hand, contains the
best weight obtained for the dataset with this architecture
Required arguments:
-a --architecture Which architecture will be used to extract the features (Xception|VGG16|VGG19|ResNet50|InceptionV3|InceptionResNetV2|MobileNet|MobileNetV2|DenseNet121|DenseNet201|NASNetMobile|NASNetLarge|EfficientNetV2L)
Optional arguments:
| Parameter | Default | Description |
|---|---|---|
| -p --path | dtd/images/ | Path where the images are found |
| -s --seed | 1994 | |
| -hi -–height_image | 300 | |
| -wi --width_image | 300 | |
| -qib --quantity_images_batch | 30 | |
| -–patience | 3 | |
| -e --epochs | 25 | |
| -lr --learning_rate | 1e-04 |