OpenPath Corpus
Public-only whole-slide histopathology tile corpus used to pre-train OpenPath, a ViT-g/14 pathology foundation model. Tiles are re-extracted at native ~40× (0.5 µm-per-pixel), which is the data lever that lets OpenPath rank #1 on the contamination-free AMC-HCC-ST benchmark among seven foundation models (see the code / model card).
Contents
- 33,991 WebDataset shards (
.tar), ~17 TB, ~834 M tiles, 224×224 at native 40×. - Seven public sources (sharded into slots such as
tcga_s0…s5):
| Source | Shards | License | Notes |
|---|---|---|---|
TCGA (tcga_s0…s5) | 20,589 | NIH / GDC open-access tier | attribution required (see below) |
GTEx (gtex_s0…s1) | 4,965 | GTEx Portal permissive (attribution-only); CC-BY 4.0 DICOM mirror available | attribution required |
TCIA (tcia_s0…s1) | 3,891 | CC-BY 3.0 / 4.0 (collection-specific) | pathology collections |
| ACROBAT | 3,318 | CC-BY 4.0 | breast H&E + IHC |
| CAMELYON16/17 | 784 | CC0 (public domain) | breast lymph node |
| SurGen | 426 | CC-BY 4.0 | colorectal |
| MIDOG (2021 / 2022 / ++) | 18 | CC-BY 4.0 | multi-tumor |
| Total | 33,991 | mixed public / redistributable | commercial use permitted |
All sources are commercial-use-permitted (CC-BY / CC0 / NIH-open). This corpus does not include any non-commercial (NC) data. PCam / CAMELYON overlap standard patch benchmarks and are therefore excluded from OpenPath's evaluation.
Format & loading
- Layout:
<source>/tiles/shards/w<id>/*.tar(WebDataset). Each tar entry is a JPEG tile with a JSON sidecar (wsi_id, tile coordinates, magnitude). - Training glob:
*/tiles/shards/w*/*.tar.
import webdataset as wds
url = "path/to/openpath-corpus/tcga_s0/tiles/shards/w0/000000.tar"
ds = wds.WebDataset(url).decode("pil").to_tuple("jpg", "json")
for img, meta in ds:
... # img: 224×224 PIL RGB, meta: {wsi_id, x, y, mag, ...}
Access & terms (gated)
Gated, manual approval. For research use. When you use the corpus, cite the OpenPath paper and preserve each source's attribution, e.g.:
The results shown here are in whole or part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.
Data were obtained from the GTEx Portal (and/or dbGaP accession phs000424).
Related artifacts
| Artifact | Hugging Face repo | Notes |
|---|---|---|
| Corpus | taejoon89/openpath-corpus | This repository |
| Checkpoints | taejoon89/openpath-checkpoints | teacher checkpoints (training_0 … training_345000); released = training_316250 |
| Code | taejoon89/openpath | training & evaluation code (also on GitHub) |
Citation
@misc{openpath2026,
title = {OpenPath: Public-Data Pathology Foundation Models and Leakage-Free Evaluation},
author = {Tae Joon Jun},
year = {2026},
note = {https://huggingface.co/taejoon89/openpath}
}
Acknowledgements
This research was supported by a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: HR21C0198); the Advanced GPU Utilization Support Program funded by the Government of the Republic of Korea, Ministry of Science and ICT; and the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (grant number: RS-2026-25522634).
License
The corpus redistributes public pathology datasets under their original CC-BY / CC0 / NIH-open terms (all redistributable, commercial use permitted). See the per-source table above and cite/attribute each source accordingly.