Hal1010/HROCT-Bench
收藏Hugging Face2026-04-20 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/Hal1010/HROCT-Bench
下载链接
链接失效反馈官方服务：
资源简介：
---
pretty_name: HROCT-Bench
task_categories:
- image-segmentation
license: other
---
# HROCT-Bench
## Dataset Description
**HROCT-Bench** is a hierarchical benchmark for retinal OCT layer segmentation under heterogeneous annotation granularities. It is designed to support research on cross-dataset, cross-granularity, and universal retinal layer segmentation.
This public release contains the subsets of HROCT-Bench that can be redistributed openly. The currently included public datasets are:
- **HC-MS**
- **A2A-SDOCT**
- **DUKE DME**
- **DUKE AMD**
- **HEG**
- **OIMHS**
In the full HROCT-Bench benchmark, we also use several additional datasets that are **not included in this release** due to restrictions imposed by the original dataset owners. These restricted datasets include:
- **OCTA-500**: https://ieee-dataport.org/open-access/octa-500
- **GCN**: https://yuyeling.com/project/mgu-net/
- **GOALS**: https://aistudio.baidu.com/competition
- **NR206**: https://github.com/Medical-Image-Analysis/NR206-Dataset/tree/main/dataset
- **Harvard-EF30K**: https://github.com/osamakhaan/Harvard-EyeFairness
For these restricted subsets, we provide the original access links above. Users should obtain the data directly from the original sources in accordance with their access policies. We also provide preprocessing scripts so that users can process the original data and obtain the preprocessed version used in our benchmark.
This release is intended to facilitate open and reproducible research while respecting the data-sharing policies of the original dataset providers.
## License
**License: other**
The data released in this repository are curated and preprocessed from multiple publicly accessible source datasets. The use of each subset remains subject to the license terms, access conditions, and redistribution policies of the corresponding original dataset provider.
Users are responsible for reviewing and complying with the original terms before downloading or using the data. For restricted subsets, we only provide the original access links and preprocessing scripts, and users must obtain the raw data directly from the original sources.
This repository does **not** claim to relicense the underlying raw data beyond what is permitted by the original data providers.
提供机构：
Hal1010
