Angelou0516/Endovis2017

Medical Image Segmentation Dataset

Dataset Description

This dataset contains medical images and segmentation masks, optimized for SAM/SAM2 evaluation.

Dataset Statistics

• Splits: train, val
• Total Samples: 2701
• Format: Parquet (HuggingFace Datasets format)

Usage

Load with 🤗 Datasets

from datasets import load_dataset

# Load dataset
dataset = load_dataset("Angelou0516/Endovis2017")

# Access samples
for split in dataset.keys():
    print(f"{split}: {len(dataset[split])} samples")

# Get a sample
sample = dataset['train'][0]
image = sample['image']  # PIL Image
mask = sample['label']   # PIL Image or numpy array

Use with EasyMedSeg

# The dataset can be loaded directly with appropriate dataset class
# Example for Endovis2017:
from dataloader.image.endovis2017_dataset import Endovis2017Dataset

dataset = Endovis2017Dataset(
    mode='download',
    split='train',
    hf_repo_id='Angelou0516/Endovis2017'
)

Evaluation with SAM/SAM2

# Create config file with dataset repo
# configs/my_config.yaml:
# test_dataset:
#   DatasetName: Angelou0516/Endovis2017

# Run evaluation
python scripts/evaluation.py --config configs/my_config.yaml

Dataset Structure

Each sample typically contains:

• image: RGB image (PIL Image)
• label or mask: Segmentation mask
• image_id: Unique identifier
• file_name: Original filename
• Additional metadata fields

Citation

Please cite the original dataset and SAM/SAM2 papers when using this dataset.

@article{kirillov2023segment,
  title={Segment Anything},
  author={Kirillov, Alexander and others},
  journal={ICCV},
  year={2023}
}

@article{ravi2024sam2,
  title={SAM 2: Segment Anything in Images and Videos},
  author={Ravi, Nikhila and others},
  journal={arXiv preprint arXiv:2408.00714},
  year={2024}
}

Links

• 🤖 EasyMedSeg Framework
• 📝 Dataset Card