3D state prediction for surface vessels using IMU data and images - a deep learning approach

2022-2023 Thesis by Lance De Waele

Supervisors: Prof. dr. ir. Hiep Luong, Prof. dr. ir. Jan Aelterman
Counsellors: Ir. Tien-Thanh Nguyen (Royal Military Academy), Dr. ir. Benoit Pairet (Royal Military Academy)

In co-op. with the Royal Military Academy of Belgium

Layout of repository

.
├── windows_eny.yml         # Anaconda environment for execution on windows
├── docs/      		    # Presentations, paper, worklog, workplan, images
│   └── ...          
├── code/
│   ├── 3dmodel/                # contains augmented simulation data*
│   ├── results/                # contains all results for each model
│   ├── Notebooks/              # contains source code
|   │   ├── data_loaders/              # contains .py files with classes for data loading, sequencing, splitting, utility functions, etc.
|   │   ├── model_states/              # contains binary files with state dictionaries for each trained model
|   │   ├── models/                    # contains .py files for each model and a model_provider.py to easily access them
|   │   ├── test_notebooks/            # contains notebooks for model testing 
|   │   ├── test_results/              # contains binary files with the MSEs of each model on all test sequences
|   │   ├── train_notebooks/           # contains one notebook for each model with complete training and testing functionalities 
|   │   ├── training_results/          # contains binary files with the training and validation MSE losses
|   │   ├── pr_data_analysis.ipynb            # notebook containing all code for data analysis
└── └── └── train_results_plots.ipynb         # notebook where all plots are made for training and validation loss

*full simulation dataset can be downloaded here: https://drive.google.com/drive/folders/1RF8_wFfcIM0GIklXflPYv-tK3uaEWSSZ

Execution of the code

To execute the notebooks, you can create a conda environment with the .yml file (Windows)

Alternatively: run the first cell of any training notebook and manually import all the packages that are used in the notebook in a Python 3.8 environment.

If GPU support is desired, Pytorch should be installed in the conda environment with the cuda toolkit extensions. To do this, follow the link to the PyTorch website, select your OS and cuda version, and run the provided command in the environment's terminal. https://pytorch.org/get-started/locally/ (If your GPU only supports cuda 10.2, you will need to install a older version of pytorch)