U-Bench: A Comprehensive Understanding of U-Net through 100-Variant Benchmarking

Dataset Usage Declaration

Data Source and License Statement

This dataset is provided for academic research and educational purposes only. We have made our best efforts to comply with relevant laws, regulations, and platform terms of use when collecting and organizing this data.

Disclaimer

1. Data Compliance: We have made every effort to ensure that the uploaded data complies with the terms of use and copyright requirements of various platforms.

2. Rights Protection: If you find that this dataset contains content that violates data agreements, infringes copyrights, or violates privacy rights, please contact us immediately.

3. Rapid Response: Upon receiving complaints or notifications, we will verify the claims and remove the relevant content within 24 hours after confirmation.

Contact Information

For any questions or removal requests, please contact us through the following channels:

• Email: [email protected]
• GitHub Issues: Open Issue

Removal Commitment

We hereby commit that any content proven to violate data agreements or infringe legitimate rights and interests will be processed and completely removed within 24 hours of receiving valid notification.

Usage Reminder

Users should ensure that their downloading and usage of this dataset complies with relevant laws, regulations, and third-party platform agreements. Legal liabilities arising from the use of this dataset shall be borne by the users themselves.

Dataset Usage

1. Datasets

Please put the dataset (e.g. BUSI) or your own dataset as the following architecture:

└── U-Bench
    ├── data
        ├── busi
            ├── images
            |   ├── benign (10).png
            │   ├── malignant (17).png
            │   ├── ...
            |
            └── masks
                ├── 0
                |   ├── benign (10).png
                |   ├── malignant (17).png
                |   ├── ...
        ├── your dataset
            ├── images
            |   ├── 0a7e06.png
            │   ├── ...
            |
            └── masks
                ├── 0
                |   ├── 0a7e06.png
                |   ├── ...
    ├── dataloader
    ├── models
    ├── utils
    ├── script
    ├── main.py
    └── main_multi3d.py

2. Training & Validation

# BUSI (in-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/busi --dataset_name busi
# BUSBRA (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/BUSBRA --dataset_name BUSBRA
# ISIC18 (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/isic18 --dataset_name isic18
# SkinCancer (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/uwaterlooskincancer --dataset_name uwaterlooskincancer
# Kvasir (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/Kvasir-SEG --dataset_name Kvasir-SEG
# CHASE (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/CHASEDB1 --dataset_name CHASEDB1
# DRIVE (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/DRIVE --dataset_name DRIVE
# DSB2018 (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/DSB2018 --dataset_name DSB2018
# GlaS (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/Glas --dataset_name Glas
# Monusac (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/monusac --dataset_name monusac
# Cell (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/cellnuclei --dataset_name cellnuclei
# Convidquex (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/covidquex --dataset_name covidquex
# Montgomery (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/Montgomery --dataset_name Montgomery
# DCA (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/dca1 --dataset_name dca1
# Cystoidfluid (In-domain)
python main.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/cystoidfluid --dataset_name cystoidfluid
# Synapse (3D-Slice)
python main_multi3d.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/synapse --dataset_name synapse --num_classes 9 --input_channel 3 --val_interval 100
# ACDC (3D-Slice)
python main_multi3d.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/ACDC--dataset_name ACDC --num_classes 4 --input_channel 3 --val_interval 100

3. Zero-shot

# BUSI -> BUS (zero-shot)
python inference_case.py --max_epochs 300 --gpu 0 --batch_size 8 --model U_Net --base_dir ./data/busi --dataset_name busi --zero_shot_base_dir ./data/bus --zero_shot_dataset_name bus --just_for_test True

Citation

If using this work (dataset, weights, or benchmark results), please cite:

@article{tang2025u,
  title={U-Bench: A Comprehensive Understanding of U-Net through 100-Variant Benchmarking},
  author={Tang, Fenghe and Dong, Chengqi and Ma, Wenxin and Xu, Zikang and Zhu, Heqin and Jiang, Zihang and Wang, Rongsheng and Wang, Yuhao and Wu, Chenxu and Zhou, Shaohua Kevin},
  journal={arXiv preprint arXiv:2510.07041},
  year={2025}
}

---

Last Updated: [Date]