AAI3001 Computer Vision and Deep Learning Project

This project's overall objective is the prediction of dominant land use from Zenodo satellite images from EuroSAT dataset. The structured approach used includes data preprocessing and splitting, training the data model and validation of results. The models are then evaluated for their performances to ensure accuracy in their predictions. There are a total of 5 main python files used. The utils.py file is a shared module between Task 1 and Task 2.

Dependencies

• Matplotlib
• Numpy
• OpenCV (cv2)
• Python
• Pillow (PIL)
• PyTorch
• scikit-learn
• torchvision

File Directory

For Dataset

AAI3001_SmallProj
├── EuroSAT_RGB

Full Directory

AAI3001_SmallProj
├── EuroSAT_RGB
│   ├── AnnualCrop
│   ├── Forest
│   ├── HerbaceousVegetation
│   ├── Highway
│   ├── Industrial
│   ├── Pasture
│   ├── PermanentCrop
│   ├── Residential
│   ├── River
│   ├── SeaLake
├── Old Codes
│   ├── task1.py
│   ├── task2.py
├── AA13001 - Small Project.pdf
├── README.md
├── image_resnet50_model.pth
├── image_resnet50_multiLabel_model.pth
├── Task 1_Training and Validation Loss Curves.png
├── Task 2_Training and Validation Loss Curves.png
├── task1train.py
├── task1validate.py
├── task2train.py
├── task2validate.py
└── utils.py

Python Scripts/Files

• utils.py
  • holds the common functions for Tasks 1 and 2.
  • not required to run, the respective scripts for Tasks 1 and 2 will call this utils script.

Task 1

• task1train.py
• task1validate.py

Task 2

• task2train.py
• task2validate.py

Code Execution

• To execute the Python scripts for the respective tasks, simply open your command prompt or terminal, navigate to the "AAI3001_SmallProj" directory and run the following commands:
  • python task1train.py to train the Task 1 model.
  • python task1validate.py to validate the Task 1 model.
  • python task2train.py to train the Task 2 model.
  • python task2validate.py to validate the Task 2 model.