Initial PHOEBI release: PHOEBI-6 (40 combos, 6 species) + PHOEBI-4 (14 combos, 4 species)

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+splits.json filter=lfs diff=lfs merge=lfs -text

HOSTING.md

@@ -0,0 +1,129 @@
+# PHOEBI — Dataset Hosting Plan
+
+This is a working document for the dataset hosting decision and the
+double-blind review path. It is not part of the public release. Delete or
+update before camera-ready.
+
+## What NeurIPS 2026 D&B requires
+
+Per <https://neurips.cc/Conferences/2026/EvaluationsDatasetsHosting>:
+
+1. **Croissant metadata is mandatory.** Core fields plus the RAI extension.
+   Both are present in `croissant.json` and must be re-validated after the
+   dataset is uploaded (URL placeholders + SHA-256 hashes need filling in).
+2. **Reviewers must be able to access the dataset via a URL** before
+   acceptance.
+3. **Dataset must be publicly available by the camera-ready deadline.**
+   Failure means removal from the conference.
+4. **Four preferred hosts** (each auto-generates a backup Croissant file):
+   Hugging Face, Kaggle, Harvard Dataverse, OpenML.
+5. Self-hosting is permitted with a manual Croissant file.
+
+## Dataset size (current)
+
+| Component | Approx. size |
+|-----------|---|
+| PHOEBI-6 images (`images.tar.gz`)        | 13 GB |
+| PHOEBI-4 images (`images_4class.tar.gz`) | 2.3 GB |
+| **Total**                                | **~15 GB** |
+
+## Hosting options compared
+
+| Platform | DOI | Free at 15 GB | Auto-Croissant | Anonymity flow |
+|---|---|---|---|---|
+| **Hugging Face Datasets** | yes (community) | yes (Git LFS) | yes | throwaway HF account; profile blank |
+| **Harvard Dataverse**     | yes              | yes            | yes | explicit "Private URL" for peer review (no account needed by reviewer) |
+| **Kaggle**                | yes              | yes            | yes | private dataset + share link; account profile must be blanked |
+| **OpenML**                | no               | yes            | yes | tabular-leaning; not a great fit for image archives at this size |
+| **Zenodo**                | yes              | yes (≤50 GB)   | no  | "Restricted access" + reviewer token; manual Croissant required |
+
+## Recommendation
+
+**Primary: Hugging Face Datasets** with a throwaway account during review,
+transferred to the real lab account on acceptance. Rationale:
+
+- The four preferred hosts are equally acceptable to NeurIPS, but HF is the
+  best fit for the ML community (`datasets` library access, easy mirroring).
+- Auto-generated Croissant satisfies the requirement; our manual one in
+  `croissant.json` is the canonical version we ship inside the supplementary
+  PDF and as a top-level file in the dataset repo.
+- Throwaway-account anonymity is workable: create `AnonymousResearchTiger`
+  (no name, no avatar, no bio), upload, do not cross-link to anything else.
+
+**Fallback: Harvard Dataverse** if reviewers raise an anonymity concern with
+the throwaway account. Dataverse has a purpose-built "Private URL" feature
+that reviewers can use without registering.
+
+**Skip:** Kaggle (account profile leakage harder to scrub), OpenML (not
+designed for image archives at this size), Google Drive / Dropbox (no DOI,
+not a preferred host, will likely fail desk review).
+
+## Step-by-step (Hugging Face path)
+
+1. **Create a throwaway HF account**
+   - Username: e.g. `AnonymousResearchTiger`.
+   - Email: a fresh address that does not identify the lab.
+   - Profile: leave blank (no name, no avatar, no bio, no organisation).
+   - Do NOT enable any cross-linking (no GitHub, no website, no Twitter).
+
+2. **Build the release archives** (locally, ~15 GB free):
+   ```bash
+   python tools/build_release_archives.py
+   ```
+   Produces `data/release/{images,images_4class}.tar.gz` plus a copy of
+   `splits.json` (PHOEBI-6) and `splits_4class.json` (PHOEBI-4).
+
+3. **Re-emit the Croissant file with real hashes**:
+   ```bash
+   python tools/build_croissant.py \
+     --hosting-url-base https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main \
+     --release-dir data/release
+   ```
+   This rewrites `data/release/croissant.json` with the real `sha256` values
+   for every `FileObject` and the actual `contentUrl` values.
+
+4. **Upload to Hugging Face**:
+   ```bash
+   pip install huggingface_hub
+   hf auth login                    # use the throwaway account
+   hf upload AnonymousResearchTiger/PHOEBI data/release . --repo-type=dataset
+   ```
+   HF will auto-generate its own Croissant file at
+   `https://huggingface.co/api/datasets/AnonymousResearchTiger/PHOEBI/croissant`.
+   Both it and our hand-crafted `croissant.json` will be present.
+
+5. **Verify reviewer access**: open the dataset URL in a private browser
+   window (signed out). The page must load without prompting for sign-in.
+
+6. **Cite the dataset URL** in the paper's "Data availability" section in
+   the supplementary appendix, plus include `croissant.json` as a separate
+   supplementary file alongside the PDF (this is the artifact NeurIPS asks
+   for in the submission form).
+
+7. **On acceptance**: transfer ownership to the lab account, fill in the
+   `creator` and `publisher` fields in `croissant.json`, and re-upload.
+
+## Citing inside the paper
+
+Add the following to the supplementary "Data availability" subsection
+once the URL is final:
+
+> The PHOEBI dataset (PHOEBI-6: ~120,000 images; PHOEBI-4: ~14,000 images;
+> ~15 GB total) is hosted at `<final URL>` under the CC BY 4.0 licence.
+> The Croissant 1.0 metadata file is at `<URL>/croissant.json`. The release
+> includes the leave-combinations-out splits at seed 1337 in `splits.json`
+> (PHOEBI-6) and the random 80/10/10 split in `splits_4class.json` (PHOEBI-4).
+
+## Checklist before submission
+
+- [ ] Throwaway HF account created with profile fully blanked.
+- [ ] `tools/build_release_archives.py` run; all archives present in
+      `data/release/`.
+- [ ] `tools/build_croissant.py` run; `croissant.json` has no
+      PLACEHOLDER values left.
+- [ ] HF upload complete; dataset visible to a signed-out user.
+- [ ] Auto-generated HF Croissant present at `/croissant` endpoint.
+- [ ] `croissant.json` validated locally with `mlcroissant validate
+      data/release/croissant.json`.
+- [ ] URL added to supplementary "Data availability" section.
+- [ ] `croissant.json` attached as a separate file in the submission form.

LICENSE

@@ -0,0 +1,40 @@
+Creative Commons Attribution 4.0 International (CC BY 4.0)
+
+The BIDS dataset is licensed under the Creative Commons Attribution 4.0
+International License (CC BY 4.0). To view a copy of this license, visit
+https://creativecommons.org/licenses/by/4.0/ or send a letter to Creative
+Commons, PO Box 1866, Mountain View, CA 94042, USA.
+
+You are free to:
+  Share — copy and redistribute the material in any medium or format
+  Adapt — remix, transform, and build upon the material for any purpose,
+          even commercially.
+
+Under the following terms:
+  Attribution — You must give appropriate credit, provide a link to the
+  license, and indicate if changes were made. You may do so in any
+  reasonable manner, but not in any way that suggests the licensor endorses
+  you or your use.
+
+No additional restrictions — You may not apply legal terms or technological
+measures that legally restrict others from doing anything the license
+permits.
+
+Notices:
+  You do not have to comply with the license for elements of the material in
+  the public domain or where your use is permitted by an applicable
+  exception or limitation.
+
+  No warranties are given. The license may not give you all of the
+  permissions necessary for your intended use. For example, other rights
+  such as publicity, privacy, or moral rights may limit how you use the
+  material.
+
+When citing this dataset, please use:
+
+@inproceedings{bids2026,
+  title={BIDS: A Compositional Benchmark for Open-World Bacterial Identification},
+  author={Anonymous Authors},
+  booktitle={Advances in Neural Information Processing Systems Datasets and Benchmarks Track},
+  year={2026}
+}

README.md

@@ -0,0 +1,155 @@
+---
+license: cc-by-4.0
+pretty_name: "PHOEBI"
+language:
+- en
+task_categories:
+- image-classification
+task_ids:
+- multi-label-image-classification
+size_categories:
+- 100K<n<1M
+tags:
+- bacteria
+- microscopy
+- phase-contrast
+- multi-label-classification
+- compositional-generalization
+- open-set-recognition
+- novel-class-discovery
+- fine-grained-recognition
+- benchmark
+---
+
+# PHOEBI
+
+**P**hase-contrast **O**ptical b**E**nchmark for **B**acterial **I**dentification — a benchmark and framework for open-world identification of mixed bacterial cultures from optical phase-contrast microscopy.
+
+## At a glance
+
+The release contains two subsets that share the same imaging pipeline:
+
+| Subset | Species | Combinations | Images |
+|--------|---------|--------------|--------|
+| **PHOEBI-6** (primary) | 6 (`bs`, `bt`, `fj`, `ka`, `mx`, `pf`) | 40 | ~120,000 |
+| **PHOEBI-4** (legacy)  | 4 (`b`, `f`, `k`, `p`)                 | 14 | ~14,000  |
+
+Both subsets:
+
+- 1024×1024×3 JPEG images at 1000× total magnification (100× oil-immersion, NA 1.25).
+- Random 80/10/10 split (PHOEBI-6 also ships a leave-combinations-out (LCO) split) at seed 1337.
+- Croissant 1.0 metadata in `croissant.json`.
+- Released under CC BY 4.0.
+
+The 4-class subset was collected under a separate microscopy session and is included to support cross-session replication of the compositional-collapse finding (paper §4.5, Appendix E.4).
+
+## Species (PHOEBI-6)
+
+| Code | Latin name | Gram | Motility | Cell length |
+|------|------------|------|----------|-------------|
+| `bs` | *Bacillus subtilis*       | + | peritrichous flagella | 4–10 µm |
+| `bt` | *Bacillus thermoamylovorans* | + | peritrichous flagella | ~4 µm |
+| `fj` | *Flavobacterium johnsoniae* | − | gliding | 5–10 µm |
+| `ka` | *Klebsiella aerogenes*    | − | peritrichous flagella | 1–3 µm (encapsulated) |
+| `mx` | *Myxococcus xanthus*      | − | gliding | 5–10 µm |
+| `pf` | *Pseudomonas fluorescens* | − | polar flagella | 1.5–3 µm |
+
+## Directory layout
+
+```
+.
+├── croissant.json                 # Croissant 1.0 metadata (core + RAI)
+├── LICENSE                        # CC BY 4.0
+├── README.md                      # this file
+├── splits.json                    # PHOEBI-6 random + LCO split definitions (seed 1337)
+├── images.tar.gz                  # PHOEBI-6: ~120,000 1024×1024 images
+├── splits_4class.json             # PHOEBI-4 random split definition (seed 1337)
+└── images_4class.tar.gz           # PHOEBI-4: ~14,000 1024×1024 images
+```
+
+After extraction the layout is:
+
+```
+images/                            # PHOEBI-6
+├── bs/                            # singletons
+├── bs_ka/                         # pairs (combo encoded by underscore-joined codes)
+├── bs_ka_fj/                      # triples
+└── bs_bt_mx_ka_fj_pf/             # six-species mixture
+
+images_4class/                     # PHOEBI-4
+├── b/                             # 4-class singletons
+├── b_f/                           # 4-class pairs
+├── b_f_k/                         # 4-class triples
+└── b_f_k_p/                       # 4-class quadruple
+```
+
+Each image is a 1024×1024×3 RGB JPEG. The PHOEBI-6 label vector is the multi-hot indicator over `[bs, bt, fj, ka, mx, pf]`; PHOEBI-4 over `[b, f, k, p]`. Example: `bs_ka_fj` → `[1, 0, 1, 1, 0, 0]`.
+
+## Splits
+
+`splits.json` (PHOEBI-6) and `splits_4class.json` (PHOEBI-4) define the protocols.
+
+**Random 80/10/10.** Image-level split with a fixed seed, intended for in-distribution closed-set characterisation.
+
+**Leave-combinations-out (LCO), seed 1337** *(PHOEBI-6 only)*. Holds out entire species combinations under three constraints: combination disjointness (held-out combinations never appear in train or val), species coverage (every species appears in at least one trained-on combination, so the protocol tests compositional generalization rather than novel-class detection), and order coverage (held-out set spans a range of combination orders).
+
+The PHOEBI-6 LCO held-out combinations at seed 1337 are: `bt`, `bs_pf`, `ka_fj`, `bs_mx_fj`, `bs_ka_pf`, `mx_ka_fj`, `bs_bt_ka_fj`, `bs_mx_fj_pf`, `bs_bt_mx_ka_fj_pf`.
+
+## Wet-lab protocol
+
+Cultures were inoculated from glycerol stocks into nutrient broth (8 g L⁻¹), sterilised by autoclave (121 °C, 15 min), and grown at 30 °C with orbital shaking at 250 rpm for 72–120 hours. Once each culture reached its characteristic growth stage, Petri dishes were prepared from the broth and imaged on an inverted phase-contrast microscope (100× oil-immersion, NA 1.25). For PHOEBI-6 three pairwise combinations (`bs`+`bt`, `bt`+`fj`, `ka`+`pf`) and the five-species combinations were not collected and are absent from the release.
+
+## Tasks supported
+
+1. **Multi-label species presence detection** in mixed cultures.
+2. **Compositional generalization** (LCO protocol).
+3. **Open-set rejection** (leave-one-class-out evaluation; see paper §4.4).
+4. **Novel-class discovery** (multi-label compositional NCD; see paper §4.4).
+
+## Tasks NOT supported
+
+- **Proportion / composition estimation.** Ground-truth species ratios are unavailable; growth rates differ across species and the wet-lab procedure records only species presence, not relative abundance.
+- **Clinical pathogen identification.** No stained-smear protocol; the dataset must not be used for clinical decisions without independent validation.
+
+## Loading the data
+
+A minimal Python loader using only the standard library:
+
+```python
+import json
+from PIL import Image
+import numpy as np
+
+# Assume images.tar.gz has been extracted next to splits.json
+with open("splits.json") as f:
+    splits = json.load(f)
+
+class_names = splits["class_names"]   # ['bs', 'bt', 'fj', 'ka', 'mx', 'pf']
+
+for entry in splits["splits"]["train"]:
+    img = np.array(Image.open(entry["path"]).convert("RGB"))   # 1024 x 1024 x 3
+    label = np.asarray(entry["label"], dtype=np.int64)         # 6-vector multi-hot
+    combo = entry["combo"]                                     # combination token, e.g. "bs_ka"
+    # ... your pipeline ...
+```
+
+For the LCO protocol, use the helper at `baselines.supervised_multilabel_heldout.select_heldout(seed=1337)` in the companion code release for an apples-to-apples held-out set.
+
+## Citation
+
+```bibtex
+@misc{phoebi2026,
+  title        = {{PHOEBI}: An Open-World Benchmark for Bacterial Identification in
+                  Phase-Contrast Microscopy},
+  author       = {Anonymous Authors},
+  year         = {2026},
+  note         = {Under review at the NeurIPS 2026 Datasets and Benchmarks Track},
+  howpublished = {OpenReview: \url{https://openreview.net/PLACEHOLDER}},
+}
+```
+
+(Author and institution information will be added on acceptance.)
+
+## Contact
+
+During the double-blind review period, please use the OpenReview discussion thread for the corresponding submission. Public contact information will be added on acceptance.

croissant.json

@@ -0,0 +1,447 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "equivalentProperty": {
+      "@id": "rdfs:equivalentProperty",
+      "@type": "@id"
+    },
+    "examples": {
+      "@id": "cr:examples",
+      "@type": "@json"
+    },
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileObject": "cr:fileObject",
+    "fileProperty": "cr:fileProperty",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "samplingRate": "cr:samplingRate",
+    "sc": "https://schema.org/",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform",
+    "wd": "https://www.wikidata.org/wiki/",
+    "prov": "http://www.w3.org/ns/prov#"
+  },
+  "@type": "sc:Dataset",
+  "name": "PHOEBI",
+  "alternateName": "Phase-contrast Optical bEnchmark for Bacterial Identification",
+  "description": "PHOEBI is a benchmark for open-world identification of mixed bacterial cultures from optical phase-contrast microscopy. The primary subset (PHOEBI-6) contains approximately 120,000 1024x1024 phase-contrast microscopy images at 1000x total magnification (100x oil-immersion objective, NA 1.25), drawn from 40 in-house cultures covering all 40 single, pair, triple, quadruple, and six-species combinations of six rod-shaped bacterial species: Bacillus subtilis (bs), Bacillus thermoamylovorans (bt), Flavobacterium johnsoniae (fj), Klebsiella aerogenes (ka), Myxococcus xanthus (mx), and Pseudomonas fluorescens (pf). Three pairwise combinations (bs+bt, bt+fj, ka+pf) and the five-species combinations were not collected and are absent. A secondary 4-class subset (PHOEBI-4) collected under a separate microscopy session adds approximately 14,000 1024x1024 images covering 14 combinations of four species (b, f, k, p), included to support cross-session replication of the compositional generalization finding. The release ships JPEG images only.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://creativecommons.org/licenses/by/4.0/",
+  "url": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI",
+  "version": "2.0.0",
+  "datePublished": "2026-05-15",
+  "keywords": [
+    "bacteria",
+    "phase-contrast microscopy",
+    "multi-label classification",
+    "compositional generalization",
+    "open-set recognition",
+    "novel-class discovery",
+    "fine-grained recognition"
+  ],
+  "creator": [
+    {
+      "@type": "Person",
+      "name": "Anonymous Authors (double-blind review; identities withheld until acceptance)"
+    }
+  ],
+  "publisher": {
+    "@type": "Organization",
+    "name": "Anonymous Institution (double-blind review; identity withheld until acceptance)"
+  },
+  "citeAs": "@misc{phoebi2026, title={{PHOEBI}: An Open-World Benchmark for Bacterial Identification in Phase-Contrast Microscopy}, author={Anonymous Authors}, year={2026}, note={Under review at the NeurIPS 2026 Datasets and Benchmarks Track}}",
+  "isLiveDataset": false,
+  "rai:dataLimitations": [
+    "All PHOEBI-6 images were captured on a single inverted phase-contrast microscope under controlled conditions; PHOEBI-4 was captured under a separate but similarly single-microscope session, so cross-instrument generalization is untested in the released benchmark.",
+    "For PHOEBI-6, three of the 15 possible pairwise species combinations (bs+bt, bt+fj, ka+pf) were not collected and are absent from the release. Quintuple combinations were also not collected.",
+    "Labels reflect the experimentally-defined composition encoded by the wet-lab protocol (which species were plated together) and were not re-checked by per-image visual inspection.",
+    "The dataset does not provide a stained-smear protocol and is not intended for clinical-grade pathogen identification.",
+    "The dataset does not support proportion or composition estimation; ground-truth species ratios are unavailable because growth rates differ across species and the wet-lab procedure records only the presence of each plated species, not its relative abundance."
+  ],
+  "rai:dataBiases": [
+    "All species are rod-shaped (no cocci, no filamentous types) and motile, so the dataset over-represents motile rod morphologies relative to the broader space of cultivable bacteria.",
+    "Cell-length distribution is skewed toward 1 to 10 micrometres; species with very large or very small cells are not represented.",
+    "All cultures were grown on nutrient-broth medium at 30 degrees Celsius for 72 to 120 hours and plated at characteristic growth stage; species behaviour under different media, temperatures, or growth-stage windows is not represented.",
+    "Single-instrument acquisition per subset (one phase-contrast microscope, one camera, one operator) limits diversity along the optical-conditions axis."
+  ],
+  "rai:personalSensitiveInformation": "None. The dataset contains microscopy images of bacterial cultures only. No human subjects, no personally identifiable information, no clinically-derived material.",
+  "rai:dataUseCases": [
+    "Multi-label classification of bacterial species presence in mixed cultures from phase-contrast microscopy images.",
+    "Compositional generalization benchmarking via the leave-combinations-out (LCO) protocol released alongside PHOEBI-6.",
+    "Cross-session replication of the compositional-collapse finding using PHOEBI-4 as an independent test bed.",
+    "Open-set recognition: detecting whether an image contains a species not present in training, via a leave-one-class-out (LOOCV) protocol.",
+    "Novel-class discovery: proposing prototype representations for previously unseen species in the multi-label compositional regime.",
+    "Self-supervised or transfer-learning evaluation of frozen vision encoders on a real microscopy modality."
+  ],
+  "rai:dataSocialImpact": "Positive impact: PHOEBI supports development of computer-vision tools for research microbiology in clinical, food, and environmental settings where mixed bacterial cultures are the norm. Open and free release reduces the cost of building and benchmarking such tools. Risk: the dataset is not certified for clinical pathogen identification and should not be used to make medical decisions; deploying any model trained on PHOEBI in a clinical context would require additional validation against the appropriate stained-smear protocol and regulatory review. The species included are standard non-pathogenic or low-risk laboratory strains; the dataset does not contain images of BSL-3 or BSL-4 pathogens.",
+  "rai:hasSyntheticData": false,
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "splits-json",
+      "name": "splits.json",
+      "description": "PHOEBI-6 deterministic 80/10/10 random split and leave-combinations-out split metadata.",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/splits.json",
+      "encodingFormat": "application/json",
+      "sha256": "33912ae1bd05041119dbcf4e989cd9ff25d16ab5d2f796af166cbec2055d733b"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "splits-4class-json",
+      "name": "splits_4class.json",
+      "description": "PHOEBI-4 deterministic 80/10/10 random split metadata (seed 1337).",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/splits_4class.json",
+      "encodingFormat": "application/json",
+      "sha256": "1428ee7b7544e6cf0f633a3d5cf891d81b6f7b0dadcadd9c413f126ab9156086"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "license-file",
+      "name": "LICENSE",
+      "description": "Creative Commons Attribution 4.0 International license.",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/LICENSE",
+      "encodingFormat": "text/plain",
+      "sha256": "e652e07aca4be3270b6df166fcfef8ccb300c4bce68c47e8cec1e53c7c85dbb4"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "readme-file",
+      "name": "README.md",
+      "description": "Dataset overview, structure, splits, and citation information.",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/README.md",
+      "encodingFormat": "text/markdown",
+      "sha256": "aff34bec88d96533ebb356189e43f7edce1194bb777709b74778c9ca4b0b9d3c"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "images-archive",
+      "name": "images.tar.gz",
+      "description": "PHOEBI-6: approximately 120,000 1024x1024 JPEG images, organised by species combination folder.",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/images.tar.gz",
+      "encodingFormat": "application/gzip",
+      "sha256": "b0c86dd3625c15fabd37a707edadb5a279cbc117c04af2405f772c874bb6e471",
+      "contentSize": "13624190019 B"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "images-4class-archive",
+      "name": "images_4class.tar.gz",
+      "description": "PHOEBI-4: approximately 14,000 1024x1024 JPEG images covering 14 combinations of four species (b, f, k, p) collected under a separate microscopy session.",
+      "contentUrl": "https://huggingface.co/datasets/AnonymousResearchTiger/PHOEBI/resolve/main/images_4class.tar.gz",
+      "encodingFormat": "application/gzip",
+      "sha256": "9f241c967b6125b204c5107a6fde12dc9a6d44c3c4d9510443bfd6a17ab73e24",
+      "contentSize": "2416794209 B"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "images-fileset",
+      "name": "images",
+      "description": "PHOEBI-6 1024x1024 JPEG images inside images.tar.gz, organised by species combination folder.",
+      "containedIn": {
+        "@id": "images-archive"
+      },
+      "encodingFormat": "image/jpeg",
+      "includes": "images/*/[0-9]*.jpg"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "images-4class-fileset",
+      "name": "images_4class",
+      "description": "PHOEBI-4 1024x1024 JPEG images inside images_4class.tar.gz, organised by species combination folder.",
+      "containedIn": {
+        "@id": "images-4class-archive"
+      },
+      "encodingFormat": "image/jpeg",
+      "includes": "images_4class/*/[0-9]*.jpg"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "images",
+      "name": "images",
+      "description": "PHOEBI-6: one record per image.",
+      "key": {
+        "@id": "images/path"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "images/path",
+          "name": "path",
+          "description": "Relative path inside the dataset (images/<combo>/NNNNN.jpg).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "images-fileset"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images/image",
+          "name": "image",
+          "description": "Phase-contrast microscopy image, 1024x1024x3 RGB JPEG.",
+          "dataType": "sc:ImageObject",
+          "source": {
+            "fileSet": {
+              "@id": "images-fileset"
+            },
+            "extract": {
+              "fileProperty": "content"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images/combo",
+          "name": "combo",
+          "description": "Species combination token derived from parent folder name (e.g. bs_ka_fj).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "images-fileset"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": ".*/images/([^/]+)/\\d+\\.jpg$"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images/label",
+          "name": "label",
+          "description": "Multi-hot 6-vector indicating species presence in the order [bs, bt, fj, ka, mx, pf].",
+          "dataType": "sc:Integer",
+          "repeated": true,
+          "references": {
+            "fileObject": {
+              "@id": "splits-json"
+            },
+            "extract": {
+              "jsonPath": "$.combo_to_label"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "images_4class",
+      "name": "images_4class",
+      "description": "PHOEBI-4: one record per image.",
+      "key": {
+        "@id": "images_4class/path"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "images_4class/path",
+          "name": "path",
+          "description": "Relative path (images_4class/<combo>/NNNNN.jpg).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "images-4class-fileset"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images_4class/image",
+          "name": "image",
+          "description": "Phase-contrast microscopy image, 1024x1024x3 RGB JPEG.",
+          "dataType": "sc:ImageObject",
+          "source": {
+            "fileSet": {
+              "@id": "images-4class-fileset"
+            },
+            "extract": {
+              "fileProperty": "content"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images_4class/combo",
+          "name": "combo",
+          "description": "Species combination token derived from parent folder name (e.g. b_f_k).",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "images-4class-fileset"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": ".*/images_4class/([^/]+)/\\d+\\.jpg$"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "images_4class/label",
+          "name": "label",
+          "description": "Multi-hot 4-vector indicating species presence in the order [b, f, k, p].",
+          "dataType": "sc:Integer",
+          "repeated": true,
+          "references": {
+            "fileObject": {
+              "@id": "splits-4class-json"
+            },
+            "extract": {
+              "jsonPath": "$.combo_to_label"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "species",
+      "name": "species",
+      "description": "Reference table of the six bacterial species in PHOEBI-6.",
+      "key": {
+        "@id": "species/code"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "species/code",
+          "name": "code",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "species/latin_name",
+          "name": "latin_name",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "species/gram",
+          "name": "gram",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "species/motility",
+          "name": "motility",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "species/cell_length_um_min",
+          "name": "cell_length_um_min",
+          "dataType": "sc:Float"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "species/cell_length_um_max",
+          "name": "cell_length_um_max",
+          "dataType": "sc:Float"
+        }
+      ],
+      "data": [
+        {
+          "species/code": "bs",
+          "species/latin_name": "Bacillus subtilis",
+          "species/gram": "positive",
+          "species/motility": "peritrichous flagella",
+          "species/cell_length_um_min": 4.0,
+          "species/cell_length_um_max": 10.0
+        },
+        {
+          "species/code": "bt",
+          "species/latin_name": "Bacillus thermoamylovorans",
+          "species/gram": "positive",
+          "species/motility": "peritrichous flagella",
+          "species/cell_length_um_min": 3.0,
+          "species/cell_length_um_max": 5.0
+        },
+        {
+          "species/code": "fj",
+          "species/latin_name": "Flavobacterium johnsoniae",
+          "species/gram": "negative",
+          "species/motility": "gliding",
+          "species/cell_length_um_min": 5.0,
+          "species/cell_length_um_max": 10.0
+        },
+        {
+          "species/code": "ka",
+          "species/latin_name": "Klebsiella aerogenes",
+          "species/gram": "negative",
+          "species/motility": "peritrichous flagella",
+          "species/cell_length_um_min": 1.0,
+          "species/cell_length_um_max": 3.0
+        },
+        {
+          "species/code": "mx",
+          "species/latin_name": "Myxococcus xanthus",
+          "species/gram": "negative",
+          "species/motility": "gliding",
+          "species/cell_length_um_min": 5.0,
+          "species/cell_length_um_max": 10.0
+        },
+        {
+          "species/code": "pf",
+          "species/latin_name": "Pseudomonas fluorescens",
+          "species/gram": "negative",
+          "species/motility": "polar flagella",
+          "species/cell_length_um_min": 1.5,
+          "species/cell_length_um_max": 3.0
+        }
+      ]
+    }
+  ],
+  "prov:wasDerivedFrom": [
+    "Glycerol stocks of bacterial species maintained by the contributing wet-lab.",
+    "Mixed liquid cultures grown in nutrient broth (8 g/L) at 30 degrees Celsius with orbital shaking at 250 rpm for 72 to 120 hours.",
+    "Petri-dish plates prepared from broth cultures and imaged on a 100x oil-immersion phase-contrast microscope (NA 1.25) at 1000x total magnification."
+  ],
+  "prov:wasGeneratedBy": [
+    "Microscope acquisition: phase-contrast image capture from inverted microscopes (resolution 2592x1944, three colour channels).",
+    "Label assignment: each image inherits the multi-hot label vector from its parent folder name (e.g. bs_ka_fj -> [1,0,1,1,0,0]); labels were not re-checked by per-image visual inspection.",
+    "Splits: PHOEBI-6 ships both a deterministic random 80/10/10 split and a leave-combinations-out split at seed 1337; PHOEBI-4 ships the random split only."
+  ]
+}

images.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b0c86dd3625c15fabd37a707edadb5a279cbc117c04af2405f772c874bb6e471
+size 13624190019

images_4class.tar.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f241c967b6125b204c5107a6fde12dc9a6d44c3c4d9510443bfd6a17ab73e24
+size 2416794209

splits.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33912ae1bd05041119dbcf4e989cd9ff25d16ab5d2f796af166cbec2055d733b
+size 23973340