Add files using upload-large-folder tool

README.md

@@ -0,0 +1,134 @@
+---
+license: cc-by-4.0
+task_categories:
+  - unconditional-image-generation
+  - image-feature-extraction
+tags:
+  - histopathology
+  - pathology
+  - digital-pathology
+  - whole-slide-images
+  - pixcell
+  - diffusion
+pretty_name: PixCell Training Data
+size_categories:
+  - 10M<n<100M
+configs:
+  - config_name: all
+    default: true
+    data_files:
+      - split: train
+        path: data/*/*.parquet
+  - config_name: tcga_diagnostic
+    data_files:
+      - split: train
+        path: data/tcga_diagnostic/*.parquet
+  - config_name: tcga_fresh_frozen
+    data_files:
+      - split: train
+        path: data/tcga_fresh_frozen/*.parquet
+  - config_name: cptac
+    data_files:
+      - split: train
+        path: data/cptac/*.parquet
+  - config_name: gtex
+    data_files:
+      - split: train
+        path: data/gtex/*.parquet
+  - config_name: ptrc_hgsoc
+    data_files:
+      - split: train
+        path: data/ptrc_hgsoc/*.parquet
+---
+
+# PixCell — Training Data
+
+The 1024×1024 H&E patches used to train [PixCell](https://github.com/cvlab-stonybrook/PixCell), a generative foundation model for digital histopathology, from the publicly available sources.
+
+**23,071,877 patches from 61,319 whole-slide images (~5.2 TB)**, extracted at 20× magnification. Images are the original JPEG files (quality 70) stored byte-for-byte, not re-encoded.
+
+| config | source | slides | patches |
+|---|---|---:|---:|
+| `tcga_diagnostic` | [TCGA](https://portal.gdc.cancer.gov/) FFPE diagnostic slides, 32 cancer types | 11,713 | 9,180,591 |
+| `tcga_fresh_frozen` | [TCGA](https://portal.gdc.cancer.gov/) fresh-frozen tissue slides, 32 cancer types | 18,192 | 3,680,187 |
+| `cptac` | [CPTAC](https://www.cancerimagingarchive.net/browse-collections/) (TCIA), 11 cancer types | 5,802 | 2,196,291 |
+| `gtex` | [GTEx](https://gtexportal.org/home/histologyPage) normal tissue, 40 tissue types | 25,266 | 7,870,674 |
+| `ptrc_hgsoc` | [PTRC-HGSOC](https://www.cancerimagingarchive.net/collection/ptrc-hgsoc/) (TCIA), ovarian cancer | 346 | 144,134 |
+
+The PixCell paper's full training set also includes NLST slides and an in-house Stony Brook University (SBU) cohort; those are not part of this release.
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# stream everything, or pick one config: "tcga_diagnostic", "gtex", ...
+ds = load_dataset("srikarym/PixCell-training-data", "tcga_diagnostic", split="train", streaming=True)
+ex = next(iter(ds))
+ex["image"]  # PIL.Image, 1024x1024 RGB
+```
+
+To download a single source without `datasets`:
+
+```bash
+hf download srikarym/PixCell-training-data --repo-type dataset --include "data/cptac/*" --local-dir PixCell-training-data
+```
+
+Rows are grouped by subtype and slide. Shuffle (`ds.shuffle(buffer_size=...)`) before training.
+
+## Fields
+
+| field | type | description |
+|---|---|---|
+| `image` | Image | JPEG patch; `image.path` is the original relative path `<src>/<subtype>/single_1024/<slide_id>/<col>_<row>.jpeg` |
+| `source` | string | config name |
+| `subtype` | string | cancer type (TCGA/CPTAC) or tissue type (GTEx) |
+| `slide_id` | string | slide identifier (WSI filename up to the first `.`) |
+| `col`, `row` | int32 | DeepZoom tile address at 20× |
+| `x`, `y`, `w`, `h` | int32 | patch region in **level-0 pixels** of the WSI; the patch is this region resized to 1024×1024 |
+| `objective_power` | float32 | scanner objective power from the WSI metadata (null if absent; such slides were tiled as 20×) |
+| `mpp_x` | float32 | microns per pixel at level 0, from the WSI metadata |
+| `in_test_set` | bool | patch belongs to the subset used for FID evaluation in the paper (it is also part of the training data) |
+
+`metadata/slides.parquet` has one row per slide: `source, subtype, slide_id, wsi_file, num_patches, objective_power, mpp_x, width, height`. Use `wsi_file` to locate the original slide in its source archive.
+
+## Extraction
+
+Patches were tiled with the DeepZoom tiler from [DSMIL](https://github.com/binli123/dsmil-wsi) (`deepzoom_tiler.py --tile_size 1024`: 20× base magnification, no overlap, background edge-threshold 15). To re-crop a patch from the original WSI:
+
+```python
+import openslide
+slide = openslide.OpenSlide(wsi_path)
+patch = slide.read_region((x, y), 0, (w, h)).convert("RGB").resize((1024, 1024))
+```
+
+### Using with the PixCell training code
+
+The PixCell code expects JPEG folders plus a `patch_names_all.hdf5` index. `scripts/to_pixcell_layout.py` rebuilds that layout from the downloaded shards:
+
+```bash
+python scripts/to_pixcell_layout.py --hf_dir PixCell-training-data --out /path/to/pixcell_data
+```
+
+The VAE latents and UNI-2h embeddings used for training are not included. They can be re-extracted from the images.
+
+## License and attribution
+
+This dataset is released under CC-BY-4.0. The images are derived from public data, and use is also subject to each source's terms. Please cite and acknowledge the original sources:
+
+- **TCGA**: The results here are in whole or part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.
+- **CPTAC**: National Cancer Institute Clinical Proteomic Tumor Analysis Consortium (CPTAC); cite the individual CPTAC pathology collections on [TCIA](https://www.cancerimagingarchive.net/).
+- **GTEx**: The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS.
+- **PTRC-HGSOC**: cite the collection as listed on its [TCIA page](https://www.cancerimagingarchive.net/collection/ptrc-hgsoc/) (CC BY 4.0).
+- **TCIA**: Clark K, et al. The Cancer Imaging Archive (TCIA): Maintaining and Operating a Public Information Repository. J Digit Imaging. 2013;26(6):1045–1057.
+
+## Citation
+
+```bibtex
+@article{yellapragada2025pixcell,
+  title={PixCell: A generative foundation model for digital histopathology images},
+  author={Yellapragada, Srikar and Graikos, Alexandros and Li, Zilinghan and Triaridis, Kostas and Belagali, Varun and Kapse, Saarthak and Nandi, Tarak Nath and Madduri, Ravi K and Prasanna, Prateek and Kurc, Tahsin and others},
+  journal={arXiv preprint arXiv:2506.05127},
+  year={2025}
+}
+```

data/ptrc_hgsoc/part-00000-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:08fbdb8433f7811f74dea651a17e0e145c94cbe56757aa8aa3d71b8cf70f65d9
+size 877276805

data/ptrc_hgsoc/part-00001-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:07450d09b18adbeb0d0ef5e766a1f4a56b1178a1ebf28ff59d909cb2455e2326
+size 847895162

data/ptrc_hgsoc/part-00002-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4aac0e0cea90fc59b5a388eb886b19742b55ee91314b47e58a311ca22aa0e483
+size 852877684

data/ptrc_hgsoc/part-00003-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48a7bc96fe3f004ce281498b5033766b652870b2de3c327dc65a4f16cd54acb0
+size 863242099

data/ptrc_hgsoc/part-00004-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:02e9d364e8679dc5e670dc97638bed8d4ba564d1a143dadbd87c61f632f99374
+size 899414068

data/ptrc_hgsoc/part-00005-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85fd65d8d84905108bb51f6ae010a63c6a52dca00f2f8229c1392d4c76a562b5
+size 896922515

data/ptrc_hgsoc/part-00006-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88a7f1ad032daa9a6d34262f7f39c93f38e766271bf2e538d8be3b55220b4c3e
+size 857662715

data/ptrc_hgsoc/part-00007-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:886cc850ff5e27f390c68255536b1ff726c48a3657f35d9903cc3e5c2733585f
+size 926118962

data/ptrc_hgsoc/part-00008-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1760a4fbe7084d5cc518f4eff349003250da9fc2bb1aac9402873cc152a4bea0
+size 907212457

data/ptrc_hgsoc/part-00009-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99f4dac851905830a2f582b408c4bdf91203a325ec950b1e06118fb3709e8633
+size 947141383

data/ptrc_hgsoc/part-00010-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3e953f56a3add0888c3975c2281abaa4696798041f51846e18d87d16f691da55
+size 933938810

data/ptrc_hgsoc/part-00011-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9efdadec959ef69148b2aa3cbcdc34242dc83f440fbda6a16c94466e8f601a34
+size 899620089

data/ptrc_hgsoc/part-00012-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1047401464a16b26882fd0e0a682df869f631630e664008afa59c6b84ab4c45d
+size 874305659

data/ptrc_hgsoc/part-00013-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e89798d275663ec6d535a3b045563fbf55aa49570aaf6daa91bc00b518f22955
+size 909144989

data/ptrc_hgsoc/part-00014-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ea2467529934a5cf8f7f92c188cd60f864820acf1d362324ac3054621b57e6b
+size 893984721

data/ptrc_hgsoc/part-00015-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a93d9f1d0ed7ba9a1647b2669924d99e8d07820b7e63bacdc9f0e7c834e9cf6
+size 903394399

data/ptrc_hgsoc/part-00016-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c74119fc3a3a5cd46567c73763411d5d22d035a35da71743ac3b95e494cbb68
+size 849270886

data/ptrc_hgsoc/part-00017-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dfb9d3f562cfc96d6b9a63289b36afcea09e5a47ca8b85d8edbddb67cd94dc16
+size 886981853

data/ptrc_hgsoc/part-00018-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:551696be8077d459e59ded46aa722b927a5662e4d49b24bfe6c8148716918e00
+size 873140960

data/ptrc_hgsoc/part-00019-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa1f6ba4c92920cbe21fcb1674dc7f8117ab300d4db2eed7102a27a5962accf3
+size 850102763

data/ptrc_hgsoc/part-00020-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26402ea8ee7534de26bcf4c731f22853ae4d549b8fcb5277bbc69c6ef98d8e7e
+size 888573586

data/ptrc_hgsoc/part-00021-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:97b573a7d7ceb1f073ec6767d3d3294880ad8f0463978e8195bf0d3c2c7bf390
+size 913750385

data/ptrc_hgsoc/part-00022-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14eceb90ed620fec6341ec1e4eaa1b67fe03d36eb261e4dbc7d7eced19acd85d
+size 867324801

data/ptrc_hgsoc/part-00023-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:826126e4c34a1e32d193a1f3bd10a819e6074a186ec63945f4e1075d85e771e1
+size 828638967

data/ptrc_hgsoc/part-00024-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ca35445d66edad2e532140842fc6e576368ef5073d9dd9812d381a6474e1bae
+size 839666673

data/ptrc_hgsoc/part-00025-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18062b1a8881c8c1d76d847394395cc3cae0c8115633d62b588361047416a7e5
+size 861461329

data/ptrc_hgsoc/part-00026-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f14bcc8b057ef63ad8e4df39bf13659d665dfb95ae9f493a1d7833cb42d45bc9
+size 877960115

data/ptrc_hgsoc/part-00027-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d81a80b8b64fbd8f1eaf9c6b3caa2525315c8c936e07fb2d4d657c41400b2a6
+size 920715901

data/ptrc_hgsoc/part-00028-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7cb2d04555ea58932c4ce0462dd2f5ededcbefa4df073561e9bb8719c3df6166
+size 852638644

data/ptrc_hgsoc/part-00029-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3aaf50d515e02cec832908043b417c9f42cfb5cb96195311c2cea5206220be82
+size 848618111

data/ptrc_hgsoc/part-00030-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb4a409b4071a6162c5ec9aa3af0536ffe54dc335c9017bcf26ed87ea2e2b390
+size 955911371

data/ptrc_hgsoc/part-00031-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f67703ecf92d1e1cf1aa5340dcf374780167fa73e33f1c7726ddd11f8574f2d4
+size 885541796

data/ptrc_hgsoc/part-00032-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4879c70363d9b4204590f5b6e36be63e2dd44402d1c5d1b4f790ba4e213b5ee
+size 899473556

data/ptrc_hgsoc/part-00033-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0395644351ff1cd0d418c8204ac4a63e1849909c8d20ce4b6da9ffd87b14bb9e
+size 865612154

data/ptrc_hgsoc/part-00034-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7da3f976a0daf425bde02b63226ce62a9dd8ab2d5273173d7bc94ea2012246b9
+size 894340655

data/ptrc_hgsoc/part-00035-of-00036.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18df9594e5123feb55d2f54d75cf6e20c11653d3b83c54444a07b71f4df1d7c9
+size 172325351

metadata/slides.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:038e71139b1cbe6825da0095051098f24371e837eb12b85273560f707600f7ab
+size 2677885

scripts/to_pixcell_layout.py

@@ -0,0 +1,41 @@
+"""Unpack PixCell-training-data Parquet shards into the folder layout used by the PixCell training code.
+
+    python to_pixcell_layout.py --hf_dir <snapshot dir> --out <root> [--sources tcga_diagnostic gtex ...]
+
+Writes <root>/patches/<src>/<subtype>/single_1024/<slide_id>/<col>_<row>.jpeg (original JPEG bytes)
+and <root>/patches/metadata/patch_names_all.hdf5 with one key per source (e.g. "tcga_diagnostic_1024").
+Note: PTRC-HGSOC lives under "others/" (key "others_1024"), matching the original layout.
+"""
+import argparse
+from collections import defaultdict
+from pathlib import Path
+
+import h5py
+import pyarrow.parquet as pq
+from tqdm import tqdm
+
+p = argparse.ArgumentParser()
+p.add_argument("--hf_dir", type=Path, required=True)
+p.add_argument("--out", type=Path, required=True)
+p.add_argument("--sources", nargs="*", default=None)
+args = p.parse_args()
+
+patches = args.out / "patches"
+names = defaultdict(list)
+sources = args.sources or sorted(d.name for d in (args.hf_dir / "data").iterdir())
+for source in sources:
+    for f in tqdm(sorted((args.hf_dir / "data" / source).glob("*.parquet")), desc=source):
+        for batch in pq.ParquetFile(f).iter_batches(columns=["image"], batch_size=256):
+            for img in batch.column("image").to_pylist():
+                dst = patches / img["path"]
+                dst.parent.mkdir(parents=True, exist_ok=True)
+                dst.write_bytes(img["bytes"])
+                names[img["path"].split("/")[0] + "_1024"].append(img["path"])
+
+(patches / "metadata").mkdir(parents=True, exist_ok=True)
+with h5py.File(patches / "metadata/patch_names_all.hdf5", "a") as h5:
+    for key, paths in names.items():
+        if key in h5:
+            del h5[key]
+        h5.create_dataset(key, data=paths, dtype=h5py.string_dtype())
+print({k: len(v) for k, v in names.items()})