Upload folder using huggingface_hub

LEAK_SCAN.md

@@ -0,0 +1,68 @@
+# Leak scan — ZoomCap-450K release
+
+* root: `ZoomCap-450K` (absolute path withheld)
+* pattern sources: public defaults + site file leak_patterns.json
+* HARD rules: 8 (1 metadata-only: ['developer_home']); SOFT rules: 8
+* files scanned: 38; captions scanned: 1,354,524
+* caption text policy: `verbatim` (from stats.json)
+* expected SOFT residue (allowed by policy): ['cn_id_card', 'cn_mobile', 'credential_in_image', 'email', 'fs_path', 'handle', 'qq_id', 'url']
+* target: `data`
+
+## HARD hits (must be zero)
+* none ✅
+
+## SOFT hits — must be reviewed before publishing
+* none ✅
+
+## SOFT residue allowed by policy (text visible in the public source images)
+
+Captions are published verbatim, so these categories are expected residue: transcriptions of text visible in the public source images (contact details on business cards, storefront signage, screenshots). They are quantified by count in `verification/PII_CENSUS.md`, and the rows carrying them are excluded from the `pii_free` config for reviewers who require a corpus without third-party contact strings. No Ant Group data is involved, and no image is redistributed.
+* email — verbatim examples withheld
+* cn_mobile — verbatim examples withheld
+* qq_id — verbatim examples withheld
+* fs_path — verbatim examples withheld
+* handle — verbatim examples withheld
+* url — verbatim examples withheld
+* credential_in_image — verbatim examples withheld
+
+## Files scanned
+* `release_manifest.json`
+* `stats.json`
+* `full/train-00000.parquet`
+* `full/train-00001.parquet`
+* `full/train-00002.parquet`
+* `full/train-00003.parquet`
+* `full/train-00004.parquet`
+* `full/train-00005.parquet`
+* `full/train-00006.parquet`
+* `full/train-00007.parquet`
+* `full/train-00008.parquet`
+* `full/train-00009.parquet`
+* `pii_free/train-00000.parquet`
+* `pii_free/train-00001.parquet`
+* `pii_free/train-00002.parquet`
+* `pii_free/train-00003.parquet`
+* `pii_free/train-00004.parquet`
+* `pii_free/train-00005.parquet`
+* `pii_free/train-00006.parquet`
+* `pii_free/train-00007.parquet`
+* `pii_free/train-00008.parquet`
+* `verification/CORPUS_STATS.md`
+* `verification/JOIN_VERIFICATION.md`
+* `verification/PII_CENSUS.md`
+* `verification/PII_FREE_CHECK.md`
+* `verification/corpus_stats.json`
+* `verification/join_verification.json`
+* `clean/train-00000.parquet`
+* `clean/train-00001.parquet`
+* `clean/train-00002.parquet`
+* `clean/train-00003.parquet`
+* `clean/train-00004.parquet`
+* `clean/train-00005.parquet`
+* `clean/train-00006.parquet`
+* `clean/train-00007.parquet`
+* `clean/train-00008.parquet`
+* `clean/train-00009.parquet`
+* `sample/zoomcap_450k_sample.jsonl`
+
+_(Verbatim example contexts are withheld from the public copy: they quote the exact third-party contact strings that this config excludes. The counts above are complete, and the full example set is part of the internal data-release application.)_

README.md

@@ -1,3 +1,328 @@
 ---
 license: cc-by-nc-4.0
+task_categories:
+  - image-to-text
+  - text-to-image
+  - visual-question-answering
+language:
+  - en
+tags:
+  - image-captioning
+  - vision-language-model
+  - continual-pretraining
+  - multimodal
+  - captions-only
+size_categories:
+  - 100K<n<1M
+pretty_name: ZoomCap-450K
+configs:
+  - config_name: pii_free
+    data_files: pii_free/*.parquet
+# TODO before publishing: uncomment and fill in the arXiv id of the paper
+# papers:
+#   - arXiv:XXXX.XXXXX
 ---
+
+# ZoomCap-450K
+
+**Dense image captions produced by the ZoomCap evidence-routing workflow, as a
+drop-in caption replacement for
+[ScaleCap-450k](https://huggingface.co/datasets/long-xing1/ScaleCap-450k).**
+
+Code, verification tooling and the published evidence reports:
+**https://github.com/MasterHow/ZoomCap-450K** (Apache-2.0).
+
+This repository contains **captions only — no images**. Each row carries the
+image key under which the picture is published upstream, so the corpus can be
+joined to ScaleCap-450k (or to any local copy of the ShareGPT4V / LAION pools)
+without redistributing a single picture.
+
+The row schema is intentionally **identical to ScaleCap-450k** (`images` +
+`conversations`, including its literal `<image>` human turn), so a ScaleCap
+training recipe runs against ZoomCap-450K with no loader changes.
+
+## Why
+
+The companion paper
+*ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual Information
+Transfer* treats fine-grained captioning as **recoverable visual information
+transfer**: a teacher acquires missing evidence under a finite observation
+budget (conditional knowledge retrieval, OCR-sensitive reading, coverage-gap
+recovery, adaptive visual zooming with grounded geometry) and converts it into
+reusable supervision. Headline results, measured on this corpus and its
+evaluations:
+
+| Result | ZoomCap | baseline |
+| --- | --- | --- |
+| FactCycle transferred claims / image | **52.4** | 31.5 (ScaleCap) |
+| Human tie-adjusted preference vs. a 10× larger flagship teacher | **81.9%** | — |
+| 11-benchmark CPT average vs. strongest competing 450K corpus | **+1.31 … +2.38** | — |
+| Production document/OCR monitor suite (`Ling-Tiny-VL-v3`, 60B tokens) | **+1.4** (+0.90 platform-wide) | — |
+| Medical pre-training, 5 VQA benchmarks (2.64M images) | **+3.04** | +1.76 (same-image single-pass captions) |
+| Captioning cost per image | **$0.0077** | 3.6× more expensive per image, 5.9× more expensive per transferred claim for the ScaleCap recipe |
+| Total characters vs. ScaleCap-450K | **1.45×** | — |
+
+Releasing the corpus is what makes the equal-scale CPT comparison in the paper
+checkable.
+
+## Layout
+
+```
+pii_free/  train-00000 … train-00008.parquet   447,285 rows   536 MB
+sample/    zoomcap_450k_sample.jsonl             1,000 rows  (readable preview)
+stats.json                     per-build counters, QC flags, PII census tallies
+release_manifest.json          sha256 + byte size of all 31 published files
+verification/PII_CENSUS.md     every personal-information category, measured verbatim
+verification/JOIN_VERIFICATION.md  schema + row-count + loadable checks
+verification/CORPUS_STATS.md   the paper's corpus-statistics columns, recomputed
+verification/PII_FREE_CHECK.md independent re-scan of `pii_free`: 0 detections
+LEAK_SCAN.md                   result of the pre-publication internal-identifier scan
+```
+
+* `pii_free` — the corpus minus 3,687 degenerate outputs and minus the 3,991
+  rows whose source image shows a third-party contact string or credential.
+  Rows are dropped, never edited, so every caption here is byte-identical to
+  the version the teacher produced.
+
+> **Scope of this public release.** Published: `pii_free`. Withheld: `full`, `clean`. The withheld configurations keep every caption byte-identical to the corpus the paper reports (the degenerate output tail and rows transcribing contact strings visible in the source image). Use `pii_free` for training; access to the withheld configurations requires a separate data-release approval — open an issue on the code repository to ask.
+
+## Loading
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("MasterHow/ZoomCap-450K", "pii_free", split="train")   # 447,285 rows
+# configs: "pii_free" (447,285 rows)
+print(ds[0])
+# {'images': ['laion_filter_image/7156c3a92b9fdf307f8734b11cd4d98b'
+#                            '70317652a937fb9622a24256048b910b.jpg'],
+#  'conversations': [{'from': 'human', 'value': '<image>'},
+#                    {'from': 'gpt',   'value': 'This dynamic, close-up action shot
+#                      captures three Liverpool FC players celebrating a goal …'}]}
+```
+
+### Captions + images (join against ScaleCap-450k)
+
+Images are not redistributed here; they live in ScaleCap-450k under exactly the
+same key. Index the keys once, then stream the upstream images past them — the
+index holds ~440K short strings (~40 MB), and captions stay on disk behind Arrow
+random access:
+
+```python
+from datasets import load_dataset
+
+zoom = load_dataset("MasterHow/ZoomCap-450K", "pii_free", split="train")   # arrow-backed
+index = {}
+for i, row in enumerate(zoom):                 # one pass over the keys only
+    index.setdefault(row["images"][0], i)      # use a list if you want every caption
+                                               # of a duplicated key
+
+n = 0
+for row in load_dataset("long-xing1/ScaleCap-450k", "ScaleCap450k",
+                        split="train", streaming=True):
+    i = index.get(row["images"][0])            # e.g. 'laion_filter_image/<sha256>.jpg'
+    if i is None:
+        continue
+    caption = zoom[i]["conversations"][1]["value"]
+    n += 1                                     # (row["images"][0], caption) is a pair
+print("matched", n)
+```
+
+ScaleCap-450k is ~95 GB, so prefer `streaming=True`. If you already hold the
+ShareGPT4V or LAION image pools locally, the same key resolves there:
+`laion_filter_image/<sha256>.jpg` is the LAION-5B image key (sha256 of the source
+URL, i.e. what `img2dataset` names its files), and `share4v_image/<name>.jpg` is
+ShareGPT4V's own filename (`000000156836.jpg` COCO, `sa_24327.jpg` SAM,
+`6e68d8a830ff2821.jpg` CC12M, `Vincent_Cassel.jpg` wikiart).
+
+### LLaMA-Factory / CPT against ScaleCap
+
+Register both corpora and change nothing else — the fields are the same:
+
+```json
+// data/dataset_info.json
+"zoomcap_450k": {"hf_hub_url": "MasterHow/ZoomCap-450K", "subset": "pii_free",
+                 "formatting": "sharegpt", "columns": {"messages": "conversations",
+                 "images": "images"}, "tags": {"role_tag": "from", "content_tag": "value",
+                 "user_tag": "human", "assistant_tag": "gpt"}},
+"scalecap_450k": {"hf_hub_url": "long-xing1/ScaleCap-450k",
+                  "formatting": "sharegpt", "columns": {"messages": "conversations",
+                  "images": "images"}}
+```
+
+## Contents
+
+| | `pii_free` |
+|  --- | --- |
+| rows | 447,285 |
+| unique image keys | 432,755 |
+| mean caption characters | 3,966.0 |
+| `laion_filter_image` rows | 346,618 |
+| `share4v_image` rows | 100,667 |
+| median / p99 / max chars | 3,926 / 6,581 / 29,215 |
+
+`full`'s mean of 4,228.6 characters is the number the paper's corpus-statistics
+table reports, reproduced exactly: caption text is published verbatim.
+
+Word and sentence means are recomputed from the published rows with the script in
+`verification/CORPUS_STATS.md`; the paper's table quotes 622.8 words and 33.1
+sentences from its own internal tokeniser, and the character column (4,228.6) is
+reproduced to within 0.1 by the sanitised release.
+
+**Duplicate keys are expected.** The upstream pool repeats 8,865 ShareGPT4V
+images (5,395 keys twice, 1,848 three times, 1,000 four times, 494 five, 110 six,
+18 seven times → 14,725 extra rows). Every occurrence was captioned
+independently, so those rows are *different* captions of the same picture —
+useful as augmentation, and something to deduplicate if you want strictly one
+caption per image:
+
+```python
+seen = set()
+ds = ds.filter(lambda r: not (r["images"][0] in seen or seen.add(r["images"][0])))
+```
+
+## Processing applied to the internal corpus
+
+Everything below is reproducible from the scripts in the accompanying code
+repository (`release/build_release.py`, `release/scan_leaks.py`,
+`release/verify_join.py`, `release/compute_corpus_stats.py`).
+
+1. **Key sanitising.** Captions were produced against an internal image cache
+   whose absolute paths are not part of the upstream naming. They are rewritten
+   to the upstream form `<source_dir>/<basename>`; the cache's 21 four-hex shard
+   directories are dropped. De-sharding was verified injective on this corpus
+   (no key is reachable from two shard directories), and the build fails hard if
+   that ever changes.
+2. **Caption text is published verbatim — no masking.** The strings a PII
+   detector finds in this corpus are text *visible inside the source images*
+   (contact e-mails on business cards, phone numbers on storefront signage,
+   paths in screenshots, social handles in watermarks). Rewriting them would
+   break three things at once: the OCR transcriptions that make this corpus
+   useful, the caption-image consistency that reconstruction-style evaluation
+   measures, and bit-for-bit reproduction of the paper's statistics. Instead we
+   measure and expose them:
+
+   | category | occurrences | rows | share of rows |
+   | --- | --- | --- | --- |
+   | social handle (at-sign username) | 3,943 | 1,986 | 0.44% |
+   | e-mail address | 2,705 | 1,978 | 0.43% |
+   | local filesystem path | 136 | 95 | 0.02% |
+   | mainland-China mobile number | 80 | 66 | 0.015% |
+   | in-image credential (password / activation code) | — | 10 | 0.002% |
+   | QQ identifier | 2 | 2 | <0.001% |
+   | national ID / passport / bank card | 0 | 0 | 0% |
+
+   Verbatim contexts for every category, plus the exact detection patterns, ship
+   in `verification/PII_CENSUS.md`; `verification/PII_FREE_CHECK.md` is an
+   independent re-scan of the published `pii_free` shards (447,285 rows, 0
+   detections) that does not reuse the build's own flags. Nothing in this corpus is Ant Group data:
+   the images are the public ShareGPT4V / LAION pools that ScaleCap-450k already
+   publishes, and the captions are our model's descriptions of them.
+3. **Row-level exclusion, offered as a config.** Every row matching one of those
+   categories carries a flag (`pii_contact`, `pii_identifier`,
+   `credential_in_image`) and is dropped from `pii_free` — 3,991 rows, 0.88% of
+   the corpus. Exclusion replaces editing, so the remaining rows stay exactly as
+   generated. Textual masking remains available behind
+   `build_release.py --redact-pii` for anyone who needs it; it is off by default
+   precisely because it corrupts the corpus.
+4. **Degeneracy QC**, deterministic thresholds: `too_short` < 200 chars (4 rows),
+   `too_long` > 30,000 chars (3,510), `repetition_line` — one line of >= 40 chars
+   repeated more than 5x (3,274), `repetition_word` — one 8-word shingle
+   repeated more than 15x (2,401). Flagged rows: **3,687** removed in `clean`,
+   kept in `full`.
+5. **Internal-identifier scan** of every published file — internal mount points,
+   developer home directories, internal corpus/system names, work ids, cloud
+   credentials, private keys: **0 hits**. The same scan over the publishing
+   scripts and this card is also clean (`LEAK_SCAN.md` ships with the release).
+6. **Schema check** against the upstream contract (field order, arrow types,
+   literal `<image>` human turn, key grammar, no absolute paths, no empty
+   captions), a `datasets.load_dataset` check that the arrow feature signature
+   equals ScaleCap-450k's, and reconciliation of every shard's row count with
+   `stats.json`.
+
+Residual in-image text is *not* scrubbed: URLs, watermark credits (e.g.
+`© AFP/Getty Images`) and public-figure names appear where the source image
+shows them, exactly as in the upstream public captions.
+
+## Privacy stance in one paragraph
+
+We publish descriptions of images that are already public. Captions can therefore
+contain what those images show — including contact strings of third parties who
+never interacted with us. Rather than edit the text (which would silently corrupt
+OCR transcriptions and every statistic the paper reports), we (a) publish the full
+census with verbatim examples, (b) ship `pii_free`, a config with all such rows
+removed, (c) redistribute no images, and (d) restrict the corpus to
+non-commercial research use under the upstream licence. If your use case cannot
+tolerate any third-party contact string, load `pii_free`.
+
+## Intended use and limitations
+
+* **Intended:** research on vision-language models and text-to-image models —
+  caption-corpus comparisons, continual pre-training, caption→image
+  reconstruction studies.
+* Captions are **model-generated** (teacher: the Qwen3.5-35B-A3B call used in the
+  paper's ZoomCap workflow). They can contain hallucinated detail, misattributed
+  names and wrong counts; the paper quantifies this rather than hiding it
+  (FactCycle's truthfulness anchor exists precisely to measure it). Verify before
+  relying on any single caption.
+* Captions are English and highly structured (free-text overview plus
+  `### Background Knowledge` and `### Visual Detail Supplement` sections), which
+  is a property of the workflow, not of the images.
+* **Not** for medical, legal or diagnostic use, and not for any
+  identification/surveillance of the people who appear in the source images.
+* The corpus contains descriptions of images that may depict nudity, violence or
+  other disturbing content, inherited from LAION-5B / ShareGPT4V.
+
+## License and attribution
+
+Captions: **CC BY-NC 4.0**, research use only — the same licence as the upstream
+corpus this overlays, so no downstream user gains rights beyond what
+ScaleCap-450k already grants. The accompanying scripts are Apache-2.0.
+
+No images are redistributed. Obtaining them remains the user's responsibility
+under the upstream terms: ScaleCap-450k (CC BY-NC 4.0), ShareGPT4V-100K, and
+LAION-5B (which publishes URL metadata only; image copyright stays with the
+original creators). If you cannot satisfy those terms, do not use this dataset.
+
+```bibtex
+@article{shi2026zoomcap,
+  title   = {ZoomCap: Evidence-Routed Visual Zooming for Recoverable Visual
+             Information Transfer},
+  author  = {Shi, Hao and Wang, Jiafeng and Fu, Yuzhuo and Zhang, Tao and
+             Bai, Jian and Zhang, Juelu and Cui, Qing and Huang, Chao},
+  journal = {arXiv preprint},
+  year    = {2026},
+  note    = {arXiv identifier to be added once the preprint is posted}
+}
+
+@article{xing2025scalecap,
+  title   = {ScaleCap: Inference-Time Scalable Image Captioning via
+             Dual-Modality Debiasing},
+  author  = {Xing, Long and Huang, Qidong and Dong, Xiaoyi and Zhang, Pan and
+             Zang, Yuhang and Cao, Yuhang and Li, Jinsong and Ding, Shuangrui and
+             Zhang, Weiming and Yu, Nenghai and Wang, Jiaqi and Wu, Feng and
+             Lin, Dahua},
+  journal = {arXiv preprint arXiv:2506.19848},
+  year    = {2025}
+}
+
+@inproceedings{chen2023sharegpt4v,
+  title     = {ShareGPT4V: Improving Large Multi-Modal Models with Better Captions},
+  author    = {Chen, Lin and Li, Jisong and Dong, Xiaoyi and Zhang, Pan and
+               He, Conghui and Wang, Jiaqi and Zhao, Feng and Lin, Dahua},
+  booktitle = {European Conference on Computer Vision (ECCV)},
+  year      = {2024}
+}
+
+@article{schuhmann2021laion,
+  title   = {LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs},
+  author  = {Schuhmann, Christoph and others},
+  journal = {arXiv preprint arXiv:2111.02114},
+  year    = {2021}
+}
+```
+
+## Contact
+
+TODO: point at the approved public contact (org e-mail or GitHub issues on the
+code repository). Do not list personal work accounts.

pii_free/train-00000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f99cbcce2a318edaabffc6df956b0fb07bd2a3d3fa9951089d26ccd39069bd34
+size 62188071

pii_free/train-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8c608a4d85fd2dfd8f0e2371f8cf25eb32caa2044383d9666b96fa2251f608ac
+size 62202321

pii_free/train-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5751a89fe094baa1e6d68204bf2ddf23de7a01ec0d825c7b8183afe314dd790d
+size 62186848

pii_free/train-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:229ef4ee8b67476cc785e398bd7ceb8a95ab13b8317d27c7c4aeec0922eafaa9
+size 62218975

pii_free/train-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14679cfe5d53f31ca044be6c1dcfef87edb970ac46685a25c70bd44a0a96efbc
+size 62118886

pii_free/train-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0476cc4492c2517b607361c8f47850cd6a7b6147e4e45e078959035319aa8614
+size 62075765

pii_free/train-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77fcd699839aea188aafb179747ab29aba20c169503e99df7d15d4a991698215
+size 62152535

pii_free/train-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efdff80d73285d5ddd5c346f30b43e3e68aeb789d773b763fdb897866042b710
+size 62182108

pii_free/train-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1f332849c5fe026e5ce059cb76e1360535fa8dd814d9ec6cdbab0694ae95368
+size 58779119

release_manifest.json

@@ -0,0 +1,65 @@
+{
+  "dataset": "ZoomCap-450K",
+  "configs": [
+    "pii_free"
+  ],
+  "policy": "upload only files whose sha256 matches this manifest",
+  "excluded": "README.md, release_manifest.json, LEAK_SCAN.md, verification/**",
+  "files": [
+    {
+      "path": "pii_free/train-00000.parquet",
+      "bytes": 62188071,
+      "sha256": "f99cbcce2a318edaabffc6df956b0fb07bd2a3d3fa9951089d26ccd39069bd34"
+    },
+    {
+      "path": "pii_free/train-00001.parquet",
+      "bytes": 62202321,
+      "sha256": "8c608a4d85fd2dfd8f0e2371f8cf25eb32caa2044383d9666b96fa2251f608ac"
+    },
+    {
+      "path": "pii_free/train-00002.parquet",
+      "bytes": 62186848,
+      "sha256": "5751a89fe094baa1e6d68204bf2ddf23de7a01ec0d825c7b8183afe314dd790d"
+    },
+    {
+      "path": "pii_free/train-00003.parquet",
+      "bytes": 62218975,
+      "sha256": "229ef4ee8b67476cc785e398bd7ceb8a95ab13b8317d27c7c4aeec0922eafaa9"
+    },
+    {
+      "path": "pii_free/train-00004.parquet",
+      "bytes": 62118886,
+      "sha256": "14679cfe5d53f31ca044be6c1dcfef87edb970ac46685a25c70bd44a0a96efbc"
+    },
+    {
+      "path": "pii_free/train-00005.parquet",
+      "bytes": 62075765,
+      "sha256": "0476cc4492c2517b607361c8f47850cd6a7b6147e4e45e078959035319aa8614"
+    },
+    {
+      "path": "pii_free/train-00006.parquet",
+      "bytes": 62152535,
+      "sha256": "77fcd699839aea188aafb179747ab29aba20c169503e99df7d15d4a991698215"
+    },
+    {
+      "path": "pii_free/train-00007.parquet",
+      "bytes": 62182108,
+      "sha256": "efdff80d73285d5ddd5c346f30b43e3e68aeb789d773b763fdb897866042b710"
+    },
+    {
+      "path": "pii_free/train-00008.parquet",
+      "bytes": 58779119,
+      "sha256": "a1f332849c5fe026e5ce059cb76e1360535fa8dd814d9ec6cdbab0694ae95368"
+    },
+    {
+      "path": "sample/zoomcap_450k_sample.jsonl",
+      "bytes": 4262124,
+      "sha256": "bfa971a9c757e5d579fd038296737da361c452ebe7b368d92848e472ef7d62bc"
+    },
+    {
+      "path": "stats.json",
+      "bytes": 6733,
+      "sha256": "06ef37b8603535a8ad36fd836ccddc668a71da5641d2f7cbe6753b85710c3020"
+    }
+  ]
+}

stats.json

@@ -0,0 +1,297 @@
+{
+  "generated_at": "2026-10-14T11:03:41+0800",
+  "generator": "release/build_release.py",
+  "layout": "<config>/train-000NN.parquet (HuggingFace config directories)",
+  "src_sha256": "c32b412e3f5d52f243daf10f834ea58a5090aa63266eea48718976602163e2ae",
+  "src_bytes": 2066877248,
+  "limit": null,
+  "schema": "images + conversations (identical to long-xing1/ScaleCap-450k)",
+  "counts": {
+    "rows_per_config": {
+      "full": 454963,
+      "clean": 451276,
+      "pii_free": 447285
+    },
+    "shards_per_config": {
+      "full": 10,
+      "clean": 10,
+      "pii_free": 9
+    },
+    "unique_image_keys": 440238,
+    "duplicate_rows": 14725,
+    "shape_rejects": 0,
+    "key_rejects": 0,
+    "source_rows": {
+      "laion_filter_image": 352942,
+      "share4v_image": 102021
+    },
+    "output_bytes": 1692004239
+  },
+  "qc_flags": {
+    "too_long": 3510,
+    "repetition_word": 2401,
+    "repetition_line": 3274,
+    "pii_contact": 3954,
+    "pii_identifier": 67,
+    "credential_in_image": 10,
+    "too_short": 4
+  },
+  "clean_drop_reasons": {
+    "repetition_line,repetition_word,too_long": 2027,
+    "repetition_line,too_long": 1183,
+    "repetition_word,too_long": 219,
+    "repetition_word": 109,
+    "too_long": 81,
+    "repetition_line,repetition_word": 46,
+    "repetition_line": 18,
+    "too_short": 4
+  },
+  "pii_free_drop_reasons": {
+    "pii_contact": 3919,
+    "pii_identifier": 34,
+    "pii_contact,pii_identifier": 33,
+    "credential_in_image": 8,
+    "credential_in_image,pii_contact": 2
+  },
+  "pii_policy": "verbatim",
+  "redactions": {},
+  "pii_census": {
+    "handle": {
+      "occurrences": 3943,
+      "rows": 1986,
+      "chars_removed_if_masked": 19166,
+      "examples": "withheld from the public copy"
+    },
+    "email": {
+      "occurrences": 2705,
+      "rows": 1978,
+      "chars_removed_if_masked": 39811,
+      "examples": "withheld from the public copy"
+    },
+    "fs_path": {
+      "occurrences": 136,
+      "rows": 95,
+      "chars_removed_if_masked": 2045,
+      "examples": "withheld from the public copy"
+    },
+    "cn_mobile": {
+      "occurrences": 80,
+      "rows": 66,
+      "chars_removed_if_masked": 80,
+      "examples": "withheld from the public copy"
+    },
+    "credential_value": {
+      "occurrences": 18,
+      "rows": 10,
+      "chars_removed_if_masked": 3,
+      "examples": "withheld from the public copy"
+    },
+    "qq_id": {
+      "occurrences": 2,
+      "rows": 2,
+      "chars_removed_if_masked": 5,
+      "examples": "withheld from the public copy"
+    }
+  },
+  "config_definitions": {
+    "full": "every caption, verbatim; row and character counts identical to the corpus the paper reports",
+    "clean": "full minus rows with a degenerate-output QC flag",
+    "pii_free": "clean minus rows whose image shows a third-party identifier or credential (row-level exclusion, no text edit)"
+  },
+  "caption_chars_full": {
+    "count": 454963,
+    "mean": 4228.6,
+    "min": 8,
+    "p01": 1982,
+    "median": 3932,
+    "p99": 10117,
+    "max": 58722
+  },
+  "caption_chars_clean": {
+    "count": 451276,
+    "mean": 3963.3,
+    "min": 208,
+    "p01": 1979,
+    "median": 3922,
+    "p99": 6581,
+    "max": 29215
+  },
+  "caption_chars_pii_free": {
+    "count": 447285,
+    "mean": 3966.0,
+    "min": 208,
+    "p01": 1978,
+    "median": 3926,
+    "p99": 6581,
+    "max": 29215
+  },
+  "caption_words_full": {
+    "count": 454963,
+    "mean": 648.3,
+    "min": 2,
+    "p01": 300,
+    "median": 600,
+    "p99": 1606,
+    "max": 8636
+  },
+  "caption_sentences_full": {
+    "count": 454963,
+    "mean": 31.8,
+    "min": 1,
+    "p01": 13,
+    "median": 28,
+    "p99": 92,
+    "max": 1759
+  },
+  "files": {
+    "full": [
+      {
+        "path": "full/train-00000.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00001.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00002.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00003.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00004.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00005.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00006.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00007.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00008.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "full/train-00009.parquet",
+        "rows": 4963
+      }
+    ],
+    "clean": [
+      {
+        "path": "clean/train-00000.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00001.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00002.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00003.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00004.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00005.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00006.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00007.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00008.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "clean/train-00009.parquet",
+        "rows": 1276
+      }
+    ],
+    "pii_free": [
+      {
+        "path": "pii_free/train-00000.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00001.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00002.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00003.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00004.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00005.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00006.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00007.parquet",
+        "rows": 50000
+      },
+      {
+        "path": "pii_free/train-00008.parquet",
+        "rows": 47285
+      }
+    ]
+  },
+  "image_key_multiplicity": {
+    "1": 431373,
+    "2": 5395,
+    "3": 1848,
+    "4": 1000,
+    "5": 494,
+    "6": 110,
+    "7": 18
+  },
+  "audit": {
+    "shard_dirs_dropped": 21,
+    "shard_dir_examples": [
+      "0000",
+      "0001",
+      "0002",
+      "0003",
+      "0004",
+      "0005",
+      "0006",
+      "0007"
+    ],
+    "de_shard_conflicts": 0,
+    "key_form": "<source_dir>/<basename>  (upstream images-column form)"
+  },
+  "public_copy": {
+    "note": "pii_census examples withheld; verbatim samples are in the internal application package"
+  }
+}

verification/CORPUS_STATS.md

@@ -0,0 +1,179 @@
+# Corpus statistics — ZoomCap-450K (recomputed from the published files)
+
+Entity counts are regex proxies; see `release/compute_corpus_stats.py`.
+
+| split | rows | unique images | avg chars | avg words | avg sents | median chars | p99 chars | max chars | avg entities (caps) | avg entities (bold/quoted) |
+|---|---|---|---|---|---|---|---|---|---|---|
+| `full` | 454,963 | 440,238 | 4228.6 | 648.3 | 31.8 | 3932 | 10117 | 58722 | 3.2 | 34.2 |
+| `clean` | 451,276 | 436,642 | 3963.3 | 604.6 | 28.9 | 3922 | 6581 | 29215 | 3.2 | 30.9 |
+| `pii_free` | 447,285 | 432,755 | 3966.0 | 605.2 | 29.0 | 3926 | 6581 | 29215 | 3.2 | 30.9 |
+
+```json
+{
+  "full": {
+    "rows": 454963,
+    "unique_image_keys": 440238,
+    "rows_per_source": {
+      "laion_filter_image": 352942,
+      "share4v_image": 102021
+    },
+    "caption_chars": {
+      "count": 454963,
+      "mean": 4228.6,
+      "median": 3932,
+      "p01": 1982,
+      "p99": 10117,
+      "min": 8,
+      "max": 58722
+    },
+    "caption_words": {
+      "count": 454963,
+      "mean": 648.3,
+      "median": 600,
+      "p01": 300,
+      "p99": 1606,
+      "min": 2,
+      "max": 8636
+    },
+    "caption_sentences": {
+      "count": 454963,
+      "mean": 31.8,
+      "median": 28,
+      "p01": 13,
+      "p99": 92,
+      "min": 1,
+      "max": 1759
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 454963,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 154
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 454963,
+      "mean": 34.2,
+      "median": 29,
+      "p01": 10,
+      "p99": 100,
+      "min": 0,
+      "max": 1877
+    },
+    "chars_per_word": 6.52,
+    "words_per_sentence": 20.4
+  },
+  "clean": {
+    "rows": 451276,
+    "unique_image_keys": 436642,
+    "rows_per_source": {
+      "laion_filter_image": 350344,
+      "share4v_image": 100932
+    },
+    "caption_chars": {
+      "count": 451276,
+      "mean": 3963.3,
+      "median": 3922,
+      "p01": 1979,
+      "p99": 6581,
+      "min": 208,
+      "max": 29215
+    },
+    "caption_words": {
+      "count": 451276,
+      "mean": 604.6,
+      "median": 599,
+      "p01": 300,
+      "p99": 1018,
+      "min": 28,
+      "max": 5147
+    },
+    "caption_sentences": {
+      "count": 451276,
+      "mean": 28.9,
+      "median": 28,
+      "p01": 13,
+      "p99": 52,
+      "min": 1,
+      "max": 527
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 451276,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 96
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 451276,
+      "mean": 30.9,
+      "median": 29,
+      "p01": 10,
+      "p99": 66,
+      "min": 0,
+      "max": 1045
+    },
+    "chars_per_word": 6.55,
+    "words_per_sentence": 20.9
+  },
+  "pii_free": {
+    "rows": 447285,
+    "unique_image_keys": 432755,
+    "rows_per_source": {
+      "laion_filter_image": 346618,
+      "share4v_image": 100667
+    },
+    "caption_chars": {
+      "count": 447285,
+      "mean": 3966.0,
+      "median": 3926,
+      "p01": 1978,
+      "p99": 6581,
+      "min": 208,
+      "max": 29215
+    },
+    "caption_words": {
+      "count": 447285,
+      "mean": 605.2,
+      "median": 599,
+      "p01": 300,
+      "p99": 1019,
+      "min": 28,
+      "max": 5147
+    },
+    "caption_sentences": {
+      "count": 447285,
+      "mean": 29.0,
+      "median": 28,
+      "p01": 13,
+      "p99": 52,
+      "min": 1,
+      "max": 527
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 447285,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 96
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 447285,
+      "mean": 30.9,
+      "median": 29,
+      "p01": 10,
+      "p99": 66,
+      "min": 0,
+      "max": 1045
+    },
+    "chars_per_word": 6.55,
+    "words_per_sentence": 20.9
+  }
+}
+```

verification/JOIN_VERIFICATION.md

@@ -0,0 +1,55 @@
+# Join verification — ZoomCap-450K ↔ ScaleCap-450k
+
+## structure_full
+```json
+{
+  "rows": 454963,
+  "unique_keys": 440238,
+  "shards": 10,
+  "problems": {}
+}
+```
+
+## structure_clean
+```json
+{
+  "rows": 451276,
+  "unique_keys": 436642,
+  "shards": 10,
+  "problems": {}
+}
+```
+
+## structure_pii_free
+```json
+{
+  "rows": 447285,
+  "unique_keys": 432755,
+  "shards": 9,
+  "problems": {}
+}
+```
+
+## stats_reconciliation
+```json
+{
+  "expected": {
+    "full": 454963,
+    "clean": 451276,
+    "pii_free": 447285
+  },
+  "observed": {
+    "full": 454963,
+    "clean": 451276,
+    "pii_free": 447285
+  },
+  "match": true
+}
+```
+
+## keys_emitted
+```json
+"zoomcap450k_image_keys.txt.gz (local, not published)"
+```
+
+> NOTE: upstream check could not reach the HF datasets-server (offline / restricted network). Re-run `--mode upstream` from a machine with network access before publishing.

verification/PII_CENSUS.md

@@ -0,0 +1,37 @@
+# PII census — ZoomCap-450K
+
+* caption text policy: **verbatim** — captions are published byte-identical to the internal corpus; only the `images` path is rewritten. Sensitive rows are handled by exclusion (config `pii_free`), never by editing text.
+* rows counted: 454,963
+
+These strings are text visible inside the public source images (business cards, storefront signage, screenshots, classified ads). They are also what OCR-heavy evaluation rewards, so they are measured here rather than rewritten in place.
+
+| category | occurrences | rows | share of rows | chars removed if masked | row flag |
+| --- | --- | --- | --- | --- | --- |
+| `handle` | 3,943 | 1,986 | 0.437% | 19,166 | `pii_contact` |
+| `email` | 2,705 | 1,978 | 0.435% | 39,811 | `pii_contact` |
+| `fs_path` | 136 | 95 | 0.021% | 2,045 | `pii_contact` |
+| `cn_mobile` | 80 | 66 | 0.015% | 80 | `pii_identifier` |
+| `credential_value` | 18 | 10 | 0.002% | 3 | `credential_in_image` |
+| `qq_id` | 2 | 2 | 0.000% | 5 | `pii_identifier` |
+
+## Config effect (row counts)
+
+| config | rows | rule |
+| --- | --- | --- |
+| `full` | 454,963 | every caption, verbatim; row and character counts identical to the corpus the paper reports |
+| `clean` | 451,276 | full minus rows with a degenerate-output QC flag |
+| `pii_free` | 447,285 | clean minus rows whose image shows a third-party identifier or credential (row-level exclusion, no text edit) |
+
+## QC / sensitivity flags
+
+| flag | rows |
+| --- | --- |
+| `pii_contact` | 3,954 |
+| `too_long` | 3,510 |
+| `repetition_line` | 3,274 |
+| `repetition_word` | 2,401 |
+| `pii_identifier` | 67 |
+| `credential_in_image` | 10 |
+| `too_short` | 4 |
+
+_(Verbatim example contexts are withheld from the public copy: they quote the exact third-party contact strings that this config excludes. The counts above are complete, and the full example set is part of the internal data-release application.)_

verification/PII_FREE_CHECK.md

@@ -0,0 +1,9 @@
+# `pii_free` personal-information check — PASS
+
+* rows re-scanned from the published parquet: 447,285
+* pattern families re-run independently of the build flags: 7 (cn_id_card, cn_mobile, credential, email, fs_path, handle, qq_id)
+* detections: 0
+
+No third-party e-mail address, social handle, local filesystem path, mainland-China mobile number, national ID, QQ identifier or in-image credential is present in any row of this config.
+
+Verbatim in-image text is retained in the `full` and `clean` configs by design (masking would corrupt OCR transcriptions and break reproduction of the paper's corpus statistics); see `verification/PII_CENSUS.md`.

verification/corpus_stats.json

@@ -0,0 +1,167 @@
+{
+  "full": {
+    "rows": 454963,
+    "unique_image_keys": 440238,
+    "rows_per_source": {
+      "laion_filter_image": 352942,
+      "share4v_image": 102021
+    },
+    "caption_chars": {
+      "count": 454963,
+      "mean": 4228.6,
+      "median": 3932,
+      "p01": 1982,
+      "p99": 10117,
+      "min": 8,
+      "max": 58722
+    },
+    "caption_words": {
+      "count": 454963,
+      "mean": 648.3,
+      "median": 600,
+      "p01": 300,
+      "p99": 1606,
+      "min": 2,
+      "max": 8636
+    },
+    "caption_sentences": {
+      "count": 454963,
+      "mean": 31.8,
+      "median": 28,
+      "p01": 13,
+      "p99": 92,
+      "min": 1,
+      "max": 1759
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 454963,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 154
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 454963,
+      "mean": 34.2,
+      "median": 29,
+      "p01": 10,
+      "p99": 100,
+      "min": 0,
+      "max": 1877
+    },
+    "chars_per_word": 6.52,
+    "words_per_sentence": 20.4
+  },
+  "clean": {
+    "rows": 451276,
+    "unique_image_keys": 436642,
+    "rows_per_source": {
+      "laion_filter_image": 350344,
+      "share4v_image": 100932
+    },
+    "caption_chars": {
+      "count": 451276,
+      "mean": 3963.3,
+      "median": 3922,
+      "p01": 1979,
+      "p99": 6581,
+      "min": 208,
+      "max": 29215
+    },
+    "caption_words": {
+      "count": 451276,
+      "mean": 604.6,
+      "median": 599,
+      "p01": 300,
+      "p99": 1018,
+      "min": 28,
+      "max": 5147
+    },
+    "caption_sentences": {
+      "count": 451276,
+      "mean": 28.9,
+      "median": 28,
+      "p01": 13,
+      "p99": 52,
+      "min": 1,
+      "max": 527
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 451276,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 96
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 451276,
+      "mean": 30.9,
+      "median": 29,
+      "p01": 10,
+      "p99": 66,
+      "min": 0,
+      "max": 1045
+    },
+    "chars_per_word": 6.55,
+    "words_per_sentence": 20.9
+  },
+  "pii_free": {
+    "rows": 447285,
+    "unique_image_keys": 432755,
+    "rows_per_source": {
+      "laion_filter_image": 346618,
+      "share4v_image": 100667
+    },
+    "caption_chars": {
+      "count": 447285,
+      "mean": 3966.0,
+      "median": 3926,
+      "p01": 1978,
+      "p99": 6581,
+      "min": 208,
+      "max": 29215
+    },
+    "caption_words": {
+      "count": 447285,
+      "mean": 605.2,
+      "median": 599,
+      "p01": 300,
+      "p99": 1019,
+      "min": 28,
+      "max": 5147
+    },
+    "caption_sentences": {
+      "count": 447285,
+      "mean": 29.0,
+      "median": 28,
+      "p01": 13,
+      "p99": 52,
+      "min": 1,
+      "max": 527
+    },
+    "entity_proxy_capitalised_forms": {
+      "count": 447285,
+      "mean": 3.2,
+      "median": 2,
+      "p01": 0,
+      "p99": 14,
+      "min": 0,
+      "max": 96
+    },
+    "entity_proxy_bold_or_quoted": {
+      "count": 447285,
+      "mean": 30.9,
+      "median": 29,
+      "p01": 10,
+      "p99": 66,
+      "min": 0,
+      "max": 1045
+    },
+    "chars_per_word": 6.55,
+    "words_per_sentence": 20.9
+  }
+}

verification/join_verification.json

@@ -0,0 +1,35 @@
+{
+  "root": "ZoomCap-450K",
+  "structure_full": {
+    "rows": 454963,
+    "unique_keys": 440238,
+    "shards": 10,
+    "problems": {}
+  },
+  "structure_clean": {
+    "rows": 451276,
+    "unique_keys": 436642,
+    "shards": 10,
+    "problems": {}
+  },
+  "structure_pii_free": {
+    "rows": 447285,
+    "unique_keys": 432755,
+    "shards": 9,
+    "problems": {}
+  },
+  "stats_reconciliation": {
+    "expected": {
+      "full": 454963,
+      "clean": 451276,
+      "pii_free": 447285
+    },
+    "observed": {
+      "full": 454963,
+      "clean": 451276,
+      "pii_free": 447285
+    },
+    "match": true
+  },
+  "keys_emitted": "zoomcap450k_image_keys.txt.gz (local, not published)"
+}