Duke DBT detection

Official baseline for Duke DBT detection dataset

Docker setup on Linux

The folder with the data downloaded from TCIA is assumed to be /Duke-DBT.

docker build -t duke-dbt .

docker run --rm --shm-size 8G -it \
  -v /Duke-DBT:/data \
  -v `pwd`:/duke-dbt-detection \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -e DISPLAY \
  -p 8889:8889 \
  duke-dbt bash

To run jupyter notebook from the container:

jupyter notebook --allow-root --ip=0.0.0.0 --port=8889

Preprocessing

python3 preprocess.py

Training

python3 train.py

Inference

python3 inference.py --weights ./yolo.pt --predictions ./predictions.csv

Postprocessing

python3 postprocess.py --predictions ./predictions.csv --output ./predictions.csv

Evaluation

python3 evaluate.py --predictions ./predictions.csv

FROC curve from evaluation is saved as froc.png.