Machine Learning

This is the repository for the TUM Phoenix Autonomous Drive Machine Learning applications.

• Literature (TUM Phoenix Wiki)
• Todo List (Issues page)

Use on TUM Phoenix Hardware

There is a workstation available with all software preinstalled and decent training hardware (Nvidia GTX1070 8GB). Just ask the project leader for an account. Access only from TUM University network (or use LRZ-VPN).

Use on own Hardware

If you want to install it on your own device follow these instructions:

1. Install requirements using pip3 (sudo pip3 install -r requirements.txt)
2. login with your username and generate config jupyter notebook --generate-config
3. add following File Save Hook to your config:

def scrub_output_pre_save(model, **kwargs):
    """scrub output before saving notebooks"""
    # only run on notebooks
    if model['type'] != 'notebook':
        return
    # only run on nbformat v4
    if model['content']['nbformat'] != 4:
        return

    for cell in model['content']['cells']:
        if cell['cell_type'] != 'code':
            continue
        cell['outputs'] = []
        cell['execution_count'] = None

c.FileContentsManager.pre_save_hook = scrub_output_pre_save

4. Start Jupyter notebook with --config=/path/to/your/config and start editing files

Datasets

We mainly use the GTSRB Dataset. It will be loaded automatically when running the Jupyter notebooks. Additional files are on the TUM Phoenix server (please contact project leader). You may need to change the paths to your environment. Please make sure they have the same structure of GTSRB.

Structure of sign recognition folder

Don't add big files (more than 10MByte) to the git repository. Store them on the server instead.

• dicts (translation: signs <-> category number)
• utilities (utility functions and scripts for data processing, training helpers, ...)