aml-project/src at main · JulianSe26/aml-project · GitHub

Name		Name	Last commit message	Last commit date
parent directory ..
app		app
rcnn		rcnn
resnet		resnet
study		study
yolo		yolo
.dockerignore		.dockerignore
Dockerfile		Dockerfile
README.md		README.md
__init__.py		__init__.py
cocosplit.py		cocosplit.py
combo.png		combo.png
data-preparation.ipynb		data-preparation.ipynb
data-statistics.ipynb		data-statistics.ipynb
dataprep.py		dataprep.py
detection_fusion.py		detection_fusion.py
fusion_test.ipynb		fusion_test.ipynb
model-evaluations.ipynb		model-evaluations.ipynb

README.md

Covid-19 detection Kaeggle challenge source files

Folders

/app: Code for the webapp
/rcnn: Includes code to train and evaluate the Faster R-CNN.
/resnet: ResNeXt backbone training code.
/study: Study-level model.
/yolo: Yolo object detection model code

The training for all models can be started with: python train.py. They all include checkpointing ability and can be resumed at any time

Files

data-preparation.ipynb: Dataset exploration and conversion from DICOM to more usable formats
data-statistics.ipynb: Examples of class distributions of the datasets
model-evaluations.ipynb: Loss statistics and evaluation metrics of all models except the ensemble
fusion_test.ipynb: Ensemble model evaluation
Dockerfile: Dockerfile to build the webapp. Build with docker build -t <your tag here> . and run with docker run --rm -p 5000:5000 <your tag here>. Pre-built images use the tag tobiasrst/aml-project