Case Study: Satellite image object detection

The Goal of the Case Study 2 is to predict Objects like Ponds, Pools, Solar Panels and Trampolines within big Satellite images:

Requirements

---

A few additional modules are required. These modules are stored in requirements.txt and can be installed via the following command:
$ pip install -r requirements.txt

In Google Colab this modules are already installed.

File Structure

da2-case-study-2
│   README.md
│   requirements.txt
│
└───models
│   │
│   └───efficientnet_v2_best
│
└───notebooks
│   │   create_predictions.ipynb
|   |   training.ipynb
│
└───src
    │
    │
    └───d01_data
    │   │   load_data.py            <- Load train data
    |   |   save_data.py            <- Used to save the final bounding boxes with class label as csv file
    │
    └───d02_processing
    │   │   preprocess.py           <- Preprocessing of the train and val data as well as the preprocessing of the patches
    |   |   postprocess.py          <- Reduce the amount of bounding boxes
    |   |   sliding_window.py       <- Implemented the sliding window
    │
    └───d03_model
    │   │   transfer_learning.py    <- Get Preprocessing of pretrained models
    │
    └───d04_visualisation
    │   │   bounding_box.py         <- Visualize bounding boxes before and after reduction
    │
    └───d05_evalutation
    │   │   evaluate_predictions.py <- Calculate TP, FP, FN and F1-Score 

Notebooks

In the notebooks directory the create_predictions.ipynb can be used to create predictions on every image that is placed inside the data folder. Also the path of the directory can be adjusted inside of the notebook.

The training.ipynb was used to play around with the data and train different models and finally evaluate them on the 8000x8000 images.