nhannguyen2709 / rsna-breast

5th place solution for RSNA Screening Mammography Breast Cancer Detection Challenge

5th place solution for RSNA Screening Mammography Breast Cancer Detection Challenge

Repository to accompany with discussion forum post (https://www.kaggle.com/competitions/rsna-breast-cancer-detection/discussion/391979).

Pre-processing

• Download the dataset from Kaggle and unzip to data/ folder

kaggle competitions download -c rsna-breast-cancer-detection
mkdir data/
unzip rsna-breast-cancer-detection.zip -d data/

• Extract pixel arrays from DICOM files, resize with aspect-ratio preserving and save to disk

python dicom_pipeline.py data/train_images data/train_png_images --src-csv data/train.csv --size 1536

• Crop breast RoI using YOLOX detector

python crop_roi.py data/train_png_images --src-csv data/train.csv --tgt-csv data/train_kaggle_yolox.csv

Pre-training

Training and evaluation

rsna-breast's Issues

NextViT

Thank you for this great work,

I wonder which backbone did you use to get your score on Kaggle?

Have you ever tried the NextViT architecture? If so, I would appreciate it if you could share what the results were like.

Thanks