Machine Learning Course Project 2

This codebase is the implementation of the Road Segmentation Challenge of the lla_team:

Luca Multazzu
Lorenzo Brusca
Aline Janvier

Codebase description

run notebook

Jupyter Notebook to use on Google Colab

src:

network

Has the deeplab pre-trained model we use as base

cross_validation

Contains the scripts for cross validation over foreground threshold and learning rate

dataset

Contains the class of the Road Segmentation dataset

helpers

Contains a number of helper functions used throughout the codebase

model

This class has the model, with the training, test and submit methods

parameters

Has the parameters of the run

post processing

A small script for post processing

pre processing

Method that perform pre processing on the data

run

This script actually runs the code

plotting:

plot_parameters

Parameters for the plot

plotting

Plotting scripts

Also includes CSV files of the scores obtained via AI crowd

Set-Up to run on personal machine

after cloning the repository create the virtual environment with:

python3 -m pip install --user virtualenv

virtualenv -p python3 MLProject2

source MLProject2/bin/activate

python3 -m pip install -r requirements.txt

Set-Up to run on Google Colab

Open the run.ipynb notebook to Colab
Add the requirements.txt file to Colab
Load the data folders to your Google Drive
Change the data paths in the notebook to the ones in your drive
Install requirements, then restart runtime

Run on personal machine

Run with the command: cd src python3 run.py

There are four experiments we ran for the report, to choose which experiment to run use option --experiment= followed by number 1, 2. 3 or 4:

Experiment 1 runs default configurations
Experiment 2 runs with normalization on top of 1
Experiment 3 further adds data augmentation but removes normalization
Experiment 4 uses Learning Rate and Foreground Threshold found via cross-validation

You can also run the cross validation yourself with the -v option. (Note this takes a lot of time)

You can select the number of epochs to run for with option --epochs= (Default is 64)

Run on colab

After following the instructions to set up, you can choose which experiment to run in the EXPERIMENT constant (either 1, 2, 3 or 4)

You can also run cross validation by setting validation=True in the configuration cell

You can change number of epochs by changing MAX_ITER in the parameters cell