Fork of repository Deep Learning Based Damage Detection on Post-Hurricane Satellite Imagery We added additional data to try and improve detection from kaggle competition
- The dataset:
- train_another : the training data; 5000 images of each class
- validation_another: the validation data; 1000 images of each class
- test_another : the unbalanced test data; 8000/1000 images of damaged/undamaged classes
- test : the balanced test data; 1000 images of each class
Notebooks:
- exploration.ipynb - has some initial charts and data exploration steps. This was the first thing we did
- resnet50.ipynb - trying out the pretrained resnet model and retraining to our dataset with 50 epochs from hugging face
- CNN.ipynb - modified from research paper very slightly, this introduces a basic CNN model with keras
- vit.ipynb - trying to use a pretrained ViT model from HuggingFace and adapting to our dataset.
- newResnet.ipynb - some changes to the 50 epochs resnet model