[CVPR2026] LUMINA: A Multi-Vendor Mammography Benchmark with Energy Harmonization Protocol

Paper link:

arXiv: https://arxiv.org/abs/2603.14644

Dataset

Dataset on Kaggle: https://www.kaggle.com/datasets/phy710/lumina-mammography-dataset

Dataset on OSF: https://osf.io/b63jc/

Please download LUMINA_PNG, Benign_Cases.xlsx, and Malign_Cases.xlsx, then put the two xlsx files under LUMINA_PNG. Other files are provided for reference. You can create a folder called "dataset" and put LUMINA_PNG in it.

(Optional) To harmonize LUMINA dataset (generate LUMINA and LUMINA_PNG using LUMINA_RAW):

python harmonize.py --data-path PATH_TO_LUMINA_RAW -o ./LUMINA

Training and Testing

In each task (Diagnosis, BIRADS, Density), go to the corresponding folder then run

./main.sh [-model model_name] [-input_size size] [-data_path data_path]

for CNNs. Here, [-input_size] can be 224 or 512, [-model] can be efficientnet_b0, densenet121, resnet50.

./main_swin.sh [-input_size size] [-data_path data_path]

for swin-T. Other models may be supported but not tested yet. For example:

./main.sh -model efficientnet_b0 -input_size 224 -data_path /dataset/LUMINA_PNG

You can get the test results by running the command like the following:

python fold_test.py --model --data-path /dataset/LUMINA_PNG --model efficientnet_b0 --input-size 224

Here, [--input-size] can be 224 or 512, [--model] can be efficientnet_b0, densenet121, resnet50, or swin_t.

The pretrained weights are available at https://drive.google.com/drive/folders/1_40R3yt2jjLXMzLksgqo06J2KIt9-vKb?usp=sharing

Benchmark

Citation

If you use this dataset in your research, please cite our CVPR 2026 paper:

Hongyi Pan, Gorkem Durak, Halil Ertugrul Aktas, Andrea M. Bejar, Baver Tutun, Emre Uysal, Ezgi Bulbul, Mehmet Fatih Dogan, Berrin Erok, Berna Akkus Yildirim, Sukru Mehmet Erturk, Ulas Bagci. "LUMINA: A Multi-Vendor Mammography Benchmark with Energy Harmonization Protocol." CVPR 2026.