AID — SLICE-3D checkpoints

Checkpoints for AID (Adaptive Importance-guided Discretized Reconstruction), an image + tabular multimodal self-supervised pre-training framework.

Code: https://github.com/Ethan-ysliu/AID

Files

pretrain/aid_slice3d_pretrain_50bins.ckpt     pre-training checkpoint
finetune/fold{0..4}_finetune_best.pth         fine-tuned classifiers (5 folds)

Usage

huggingface-cli download EthanL06/AID-SLICE3D --local-dir ./ckpts

python run.py test \
  --checkpoint    ./ckpts/pretrain/aid_slice3d_pretrain_50bins.ckpt \
  --model_dir     ./ckpts/finetune \
  --test_csv      <test_ids.csv> \
  --metadata_csv  <metadata.csv> \
  --image_dir     <image_dir>

License

Released under CC BY-NC 4.0 — academic research use only, no commercial use.

Data: ISIC 2024 (SLICE-3D); see the ISIC archive for per-image attribution and licensing.

Research use only. Not a medical device.

Citation

@inproceedings{liu2026aid,
  title     = {Unlocking the Power of Medical Tabular Data via Semantic-Aware Multimodal Pre-training},
  author    = {Liu, Yingsheng and Li, Haiming and Zhu, Jingmin and Sun, Jiajun and
               Mar, Victoria and Janda, Monika and Soyer, H. Peter and Ge, Zongyuan and Yu, Zhen},
  booktitle = {Medical Image Computing and Computer Assisted Intervention (MICCAI)},
  year      = {2026}
}