If you clone the Git repo, you will end up with this structure below:
.
├── Dataset
│ ├── black_dress [450 entries]
│ ├── black_pants [539 entries]
│ ├── blue_dress [502 entries]
│ ├── blue_pants [512 entries]
│ ├── green_shoes [455 entries]
│ ├── red_dress [527 entries]
│ ├── red_shoes [501 entries]
│ └── white_dress [506 entries]
├── Output
├── test_examples_labels
├── logs
├── README.md
├── main.py
├── predict.py
├── train.ipynb
└── vgg.py
The different categories of the data are stored under the folder Dataset. Each category has ~450-550 entries. The trained model and the loss-accuracy plots would be saved in Output folder. The logs folder is for storing the checkpoints for monitoring the progress of loss and accuracies live on the tensorboard while the model is being trained. The folder test_examples_labels has some examples of images not in the dataset for prediction. A complete guide for running the code is available at my blog.
###Requirements
- keras == 2.2.4
- scikit-learn == 0.21.2
- pandas == 0.24.2
- numpy == 1.16.4
- opencv == 3.4.2
- matplotlib == 3.1.0
###Run Instructions To train the model, run the following command through the command line:
python3 main.py -train
To predict the class label of an unseen image, run the command
python3 main.py -predict {path_to_the_image}
The output would be displayed on the terminal and an image with name 'output.jpg' would be saved with the label and the confidence value for visualization.