Where can i drive - keras implementation

Keras Implementation of the encoder decoder network Where can I drive as probposed in the ArXiv.org article.

Getting started

This implementation uses tensorflow keras. However, no special tensorflow functions are used. So any Keras 2.0 compliant environment can be used.

Venv or virtualenv or Pipenv or system wide way (easiest)

1. Clone this git repository
2. Install reqirements.txt file in your preferred environment

pip install -r requirements.txt

Conda way (recommended)

1. Install Conda compatible environment
2. Create Conda environment and install required packages.

conda env create --file environment.yml

3. By default the CPU is used. However, it is strongly recommended to use a CUDA compatible GPU if available.

conda uninstall tensorflow && conda install tensorflow-gpu

Train the network

Envoke training via CLI:

usage: train.py [-h] [-o OUTPUT] [-ep EPOCHS] [-l LEARNING_RATE] [-b BATCH_SIZE]
                [-g GPU] [--height HEIGHT] [-w WIDTH]
                train_images train_masks val_images val_masks extension

positional arguments:
  train_images          Path to the folder containing RGB training images.
  train_masks           Path to the folder containing training masks.
  val_images            Path to the folder containing RGB validation images.
  val_masks             Path to the folder containing validation masks.
  extension             Name of the file extension. For example: '-e jpg'.
  background_pth        Path to direcotry contatining images used for background switch augmentation.  
                        For example: '/path_to/random_images/'.


optional arguments:
  -h, --help
        show this help message and exit

  -o OUTPUT, --output OUTPUT
        Path to the folder where to store model path and other training artifacts.

  -ep EPOCHS, --epochs EPOCHS
        Training epochs.

  -l LEARNING_RATE, --learning_rate LEARNING_RATE
        Learning rate for the adam solver.

  -b BATCH_SIZE, --batch_size BATCH_SIZE
        Batch size of training and validation.

  -g GPU, --gpu GPU
        Select the GPU id to train on.

  --height HEIGHT
        Height of the neural network input layer.

  -w WIDTH, --width WIDTH
        Width of the neural network input layer.

  --horizontal_flip HORIZONTAL_FLIP
        Probability of flipping image in training set. Default is 0.5.

  --brightness_contrast BRIGHTNESS_CONTRAST
        Probability of applying random brightness contrast on image in training set. Default is 0.2.

  --rotation ROTATION   
        Probability of applying random rotation on image in training set. Default is 0.9.

  --motion_blur MOTION_BLUR
        Probability of applying motion blur on image in training set. Default is 0.1.

  --background_swap BACKGROUND_SWAP
        Probability of applying background swap on image in training set. Default is 0.9.

Predict images

Envoke prediction via CLI:

usage: predict.py [-h] [-v {grayscale,heatmap,binary}] [-t THRESHOLD]
                  [-mt] [-p] [--height HEIGHT] [-w WIDTH]
                  input extension model output

positional arguments:
  input                 Path to the folder with the RGB images to be processed.
  extension             Name of the file extension. For example: <-e jpg>.
  model                 Path to the architecture/model file.
  output                Path to folder in which the segmented images are to be stored.

optional arguments:
  -h, --help            
        show this help message and exit
  
  -v {grayscale,heatmap,binary}, --vistype {grayscale,heatmap,binary}
        Visualisation type. Default is grayscale.
  
  -t THRESHOLD, --threshold THRESHOLD
        Threshold for binary classification. Default is 0.5.
  
  -mt, --multiple-thresholds
        Store all thresholds from 0-10 in 1, 10-100 in 10, 90-100 in 1 steps.
  
  -p, --progress        
        Show progress bar on stdout.
  
  --height HEIGHT
        Height of the output image.
  
  -w WIDTH, --width WIDTH
        Width of the output image.