Road Segmentation

Introduction

	Overview
Course	CS-433 Machine Learning
Year	Autumn 2020
Final Grade	5.5 / 6
Languages	Python, TensorFlow Keras
Team size	3

In the context of the EPFL CS-433 Machine Learning AICrowd challenge in which I achieved a final grade of 5.5 on a 1 to 6 scale. our goal is to create a machine learning model that labels every 16x16 patches as either road or background on satellite images from GoogleMaps. The dataset is composed of 100 training images along with their respective ground truths and the 50-image test set whose predictions are to be submitted on AICrowd. For that, we trained a U-Net model that predicts each pixel's class as well as a Convolutional Neural Network trained to output a 2D-grid of labels: one for each patch of the image. Both are implemented in TensorFlow Keras. In the end the U-Net, trained using heavy data augmentation, was more performant and is used for our final AICrowd submission, reaching an F1 of 90% score and an accuracy of 94.5%. The detailed explanations are described in this report.

Team members

• Julien Benhaim
• Hugo Lepeytre
• Julien Vignoud

Table of contents

1. Submission reproduction
2. Folder architecture
3. Training pipeline
   • Data augmentation
   • Training
   • Submission

Submission reproduction

First install the required libraries and packages used in this project. You can either install everything directly with:

pip install numpy==1.16.0
pip install -r requirements.txt

or use a virtual environment (make sure that it is not nested in another conda environment):

pip install virtualenv
virtualenv venv_tbh

For Linux and Mac OS

source venv_tbh/bin/activate

For Windows

venv_tbh\scripts\activate.bat 

Finally install the dependencies

pip install numpy==1.16.0
pip install -r requirements.txt
ipykernel install --user --name=venv_tbh

The last line creates a kernel for jupyter, in order to use the virtual environment in the notebooks. To do so, change your kernel, once the notebook running, by selecting the dropdown menu Kernel>Change kernel>venv_tbh.

This setup has been tested on a variety of machines, but there might be cases that we have not thought about. If something goes wrong, it is always possible to manually install the dependencies as described in requirements.txt. It might even work on your default Python environment, just make sure that the Tensorflow version is alright.

The setup is done and all is left is to create the predictions on the test set:

python run.py

It will load the model, run predictions on the test set and create a .csv containing a label for each patch. The submission file is created in the top folder as submission.csv. Submitting this file on AICrowd yields an F1-score of 90%, identical to our submission #109366.

Once you are done, quit the virtual environnment with

deactivate

Folder architecture

Our project contains many folders organized as follows:

project
│   README.md
│   requirements.txt
|   run.py    
|   data_augmentation.ipynb
|   pipeline.ipynb
|   report.pdf
│
└───data
│   └───train
|   |   └───original
|   |   └───generated
|   |
│   └───test
|       └───original
|       └───predictions 
│   
└───helpers
|       colab.py
|       constants.py
|       image_loading.py
│        ...
|
└───models
│       unet.py
│       cnn_patch_prediction.py
|
└───saved_models
|       best_cnn
|       final_model
|       

Here are the main subfolders description:

root

Here are the main files:

• run.py reproduces our best submission on AICrowd.
• pipeline.ipynb illustrates our training pipeline from loading data, training, validating to creating the submission file.
• data_augmentation.ipynb shows how we generated new data from the original ones.
• report.pdf explains in more detailed our methods and contributions.
• README.md
• requirements.txt

data

The data folder contains the training and test images. The first is composed of the 100 original images as well as the artificially created ones, in total around 1000 once you ran the `data_augmentation.ipynb` notebook. The test set folder contains two subfolders: the original ones with the images used for the AICrowd submission and the predictions folder, containing the predicted outputs as well as the masks superposed on the original images, to give a qualitative evaluation of our predictions.

helpers

This folder contains the python scripts used during the pipeline or to run predictions.

models

models contains two files implementing the architecture our models:

• best_cnn corresponds to the patch-wise CNN
• final_model is in fact the U-Net architecture that was used for our best submission.

saved_models

saved_models is the folder used to store the trained models using the Keras module from Tensorflow.

Training

Here is the description of the pipeline used to train the model and fine tune the hyperparameters of our final model.

• Data augmentation
• Training
• Submission

Data augmentation

According to the paper that first described the U-Net architecture [1], the performance depends heavily on data augmentation. We chose to only perform the following transformations:

• rotation
• shift
• flip
• combination of the previous transformations

The script used to generate new data is illustrated in the data_augmentation.ipynb notebook, using the Tensorflow preprocessing class ImageDataGenerator. Running this notebook will reproduce our data augmentation. The generated images can be found in the data/training/generated folder.

Training

Our models are defined in the model folder. The pipeline.ipynb notebook can be used to train a new model, using the Keras module of Tensorflow. The model used for our final submission has been trained using the data augmentated folders hard_mix, rotation, mix and flip, for a total of 900 images. The Adam optimizer was used along with the focal Tversky loss, defined in helpers/loss_functions.py, with alpha = 0.6 and gamma = 0.75. Both EarlyStopping and ReduceLROnPlateau callbacks were set on 250 epochs.

Submission

Since our model is fitted for the size of the training images (400x400) we had to come up with a method to segment the test images, that have a size of 608x608. For that we chose the following method:

1. Split the image in 4 smaller images of size 400x400
2. Segment individually each image
3. Merge the 4 masks by averaging the overlapping parts
4. Binarize the outputs between 0 and 1 and assign a label to each patch, where a patch is labeled as road is the proportion of road pixel is greater than the threshold ROAD_THRESHOLD_PATCH defined in helpers/constants.py.

This process is implemented by the function predict_submissions defined in helpers/submission.py and illustred bellow: