UNET Image Segmentation

Image Segmentation of Vehicle Driving Situations

                   

Setup virtual environment

$ cd ~
$ python3.9 -m venv <environment-name>
$ source <environment-name>/bin/activate
$ pip install -r <path-to-repo>/requirements.txt

File Structure

data/

• Images : Raw Image Data
• Labels : Labels as RGB Images
• Labels_int : Labels transfered as Integer Classes
• Test : One Test Driving Session, used to create an Video to visualize the Segmentation
• David_Description.pdf : Paper of the used Dataset

presentation/

• data_visualization : Exemplaric Images and Histogram of the Occurrence of the Labeled Classes
• evaluation : Plots to illustrate the different Results of each Model

src/

• transform_data.ipynb : Transfer RGB Labels of the Dataset into Integer Classes
• alexnet_transfer_learning.ipynb : Train Unet with Alexnet as Encoder
• resnet18_transfer_learning.ipynb : Train Unet with Resnet18 as Encoder
• shufflenet_transfer_learning.ipynb : Train Unet with Shufflenet as Encoder
• model_architecture.py : Decoder for the Resnet18
• evaluate_models.ipynb : Notebook to compare the Results of the different Models
• create_video.ipynb : Notebook to visualize the Segmentation Results as Video
• models : Already Trained Models to reload and use them without retraining

About the Authors

Johannes Mock
Simeon Grossmann

References

The used Dataset is provided by:

• https://mediatum.ub.tum.de/1596437

The Decoder used for the Resnet18 Model is taken from:

• https://github.com/jarvislabsai/siim-acr

License

GNU GENERAL PUBLIC LICENSE Version 3