Add HumanRef raw TSV training bundle and standalone loader

.gitattributes

@@ -58,3 +58,5 @@ 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
+humanref/train.annotations.tsv filter=lfs diff=lfs merge=lfs -text
+humanref/train.images.tsv filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,153 @@
+---
+language:
+- en
+task_categories:
+- image-text-to-text
+license: other
+license_name: idea-license-1.0
+license_link: https://huggingface.co/datasets/IDEA-Research/HumanRef-CoT-45k/blob/9e1afd27358d774f626a78992c7b716fa403da54/LICENSE
+size_categories:
+- 10K<n<100K
+tags:
+- visual-grounding
+- referring-expression-comprehension
+- bounding-boxes
+- tsv
+---
+
+# Grounding training data — raw TSV bundle
+
+This repository currently contains **HumanRef training data only**, repackaged
+into the indexed TSV storage layout used by
+[Rex-Omni-Finetune-ToyData](https://huggingface.co/datasets/Mountchicken/Rex-Omni-Finetune-ToyData).
+It is a derived storage export, not an official HumanRef release.
+
+| Dataset | Split | Query rows | Positive | Rejection |
+| --- | --- | ---: | ---: | ---: |
+| HumanRef | train | 45,408 | 37,057 | 8,351 |
+
+There are 86,937 selected target boxes. Images are stored as 45,408 source records;
+repeated images are retained. A prior decoded-image audit found 14,723 unique
+RGB images. This release contains no validation or test split.
+
+## Download and load
+
+Download repository files with `huggingface_hub.snapshot_download` or the Hub CLI,
+then use the included plain Python reader. This is an indexed TSV bundle, not a
+Parquet dataset or a Hugging Face `datasets.load_dataset` loading script.
+
+```bash
+pip install huggingface_hub 'Pillow>=10,<13'
+```
+
+```python
+import sys
+from huggingface_hub import snapshot_download
+
+root = snapshot_download(
+    repo_id="huuthientran/HumanRef",
+    repo_type="dataset",
+    allow_patterns=["humanref/*", "loader.py", "requirements.txt", "benchmarks.json"],
+)
+sys.path.insert(0, root)
+from loader import RawTSVDataset
+
+data = RawTSVDataset(f"{root}/humanref")
+raw = data[0]                     # original image_bytes + annotation dictionary
+sample = data.decode(raw)         # RGB image; no EXIF auto-rotation
+boxes = sample["annotation"]["boxes"]
+positive = RawTSVDataset(f"{root}/humanref", positive_only=True)
+```
+
+The loader requires only Pillow. Pin `revision` in `snapshot_download` to a
+repository commit for reproducible experiments. Data files total about 11 GiB.
+
+## Files and format
+
+```text
+README.md
+loader.py
+requirements.txt
+benchmarks.json
+humanref/
+  train.images.tsv
+  train.images.tsv.lineidx
+  train.annotations.tsv
+  train.annotations.tsv.lineidx
+  train.positive.annotations.tsv.lineidx
+  images.jsonl
+  manifest.json
+  validation.json
+  SOURCE_LICENSE
+```
+
+Image rows are `image_byte_offset<TAB>base64_original_image_bytes`.
+Annotation rows are `image_byte_offset<TAB>annotation_JSON`.
+Each `.lineidx` stores byte offsets, not row numbers. The positive index selects
+positive annotation rows without rewriting the base data. `images.jsonl` records
+image checksums and dimensions; `manifest.json` records file checksums and source
+provenance. `validation.json` reports validation performed before publication.
+
+For positive queries, `boxes` contains selected targets:
+
+```json
+{"bbox": [10, 20, 100, 200], "phrase": "the person on the left"}
+```
+
+For rejection queries, it contains a null box:
+
+```json
+{"bbox": null, "phrase": "an unmatched description"}
+```
+
+Boxes are **absolute pixel xyxy**. Source `region_map` indices select candidates
+from original pixel-xywh `gt_boxes`; only selected candidates become targets.
+Every query retains the complete original `source_annotation`, including candidate
+boxes, region mapping, and source reasoning text, together with source subset and
+offset provenance. Source reasoning is provenance, not an inference prompt or a
+direct-box answer. Separate copies of the original annotation TSVs are omitted
+because their annotations are already embedded in every exported query.
+
+Image bytes and original annotations are preserved. No resizing, re-encoding,
+EXIF transpose, clipping, normalization, quantization, filtering, or deduplication
+is applied. Qwen pixel/normalized formatting and Rex coordinate-token formatting
+belong in the experiment's runtime adapter. For a compatible grounding reader,
+configure the box format as `xyxy`.
+
+## Scope and quality
+
+All image-byte checksums, file checksums, byte offsets, selected-target mappings,
+and positive/rejection counts passed validation. Loader samples were decoded;
+full exports were checked by image checksums and headers, not a second exhaustive
+raster decode.
+
+This is raw training storage. Known duplicate queries, conflicting targets, and
+boxes extending beyond image boundaries remain unchanged. Select a training
+policy explicitly; do not silently clip or treat candidate boxes as targets.
+The manifest's `ready_for_training: false` indicates that experiment-specific
+selection and model formatting are still required, not a failed storage check.
+
+## Extending the bundle
+
+Each benchmark has its own top-level directory and manifest, using the same
+image/annotation TSV envelope. `benchmarks.json` lists only datasets actually
+included in this repository. Future releases can add `refcocog/`, `coco/`
+(COCO-Keypoints), `ap10k/`, `coco_detection/`, or `lvis/` without moving HumanRef
+or changing its loader paths. **None of those benchmarks is included yet.**
+
+Task-specific annotation schemas remain distinct: grounding, detection, and pose
+adapters must interpret their respective fields. Shared storage layout does not
+imply identical training targets. Future datasets carry their own provenance,
+validation, and license; HumanRef's license does not automatically cover them.
+
+## Source and license
+
+Source: [IDEA-Research/HumanRef-CoT-45k](https://huggingface.co/datasets/IDEA-Research/HumanRef-CoT-45k),
+revision `9e1afd27358d774f626a78992c7b716fa403da54`.
+Both `humanref_cot` and `humanref_cot_rejection` are included in source order.
+The four source TSV checksums matched the audited source revision.
+
+The upstream [IDEA License 1.0](humanref/SOURCE_LICENSE) is included unchanged;
+it grants use for non-commercial research and requires a separate license for
+commercial use. Original attribution: "T  is licensed under the IDEA License 1.0,
+Copyright (c) IDEA. All Rights Reserved."

benchmarks.json

@@ -0,0 +1,12 @@
+{
+  "format": "rex-toy-tsv",
+  "format_version": 1,
+  "datasets": {
+    "humanref": {
+      "path": "humanref",
+      "split": "train",
+      "schema": "rex-tsv-grounding-v1",
+      "manifest": "humanref/manifest.json"
+    }
+  }
+}

humanref/SOURCE_LICENSE

@@ -0,0 +1,51 @@
+IDEA License 1.0
+
+This License Agreement (as may be amended in accordance with this License Agreement, “License”), between you, or your employer or other entity (if you are entering into this agreement on behalf of your employer or other entity) (“Licensee” or “you”) and the International Digital Economy Academy (“IDEA” or “we”) applies to your use of any computer program, algorithm, source code, object code, or software that is made available by IDEA under this License (“Software”) and any specifications, manuals, documentation, and other written information provided by IDEA related to the Software (“Documentation”). 
+
+By downloading the Software or by using the Software, you agree to the terms of this License. If you do not agree to this License, then you do not have any rights to use the Software or Documentation (collectively, the “Software Products”), and you must immediately cease using the Software Products. If you are agreeing to be bound by the terms of this License on behalf of your employer or other entity, you represent and warrant to IDEA that you have full legal authority to bind your employer or such entity to this License. If you do not have the requisite authority, you may not accept the License or access the Software Products on behalf of your employer or other entity.
+
+1. LICENSE GRANT
+
+a. You are granted a non-exclusive, worldwide, transferable, sublicensable, irrevocable, royalty free and limited license under IDEA’s copyright interests to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Software solely for your non-commercial research purposes. 
+
+b. The grant of rights expressly set forth in this Section 1 (License Grant) are the complete grant of rights to you in the Software Products, and no other licenses are granted, whether by waiver, estoppel, implication, equity or otherwise. IDEA and its licensors reserve all rights not expressly granted by this License.
+
+c. If you intend to use the Software Products for any commercial purposes, you must request a license from IDEA, which IDEA may grant to you in its sole discretion.
+
+2. REDISTRIBUTION AND USE 
+
+a. If you distribute or make the Software Products, or any derivative works thereof, available to a third party, you shall provide a copy of this Agreement to such third party. 
+
+b. You must retain in all copies of the Software Products that you distribute the following attribution notice: "T  is licensed under the IDEA License 1.0, Copyright (c) IDEA. All Rights Reserved."
+
+d. Your use of the Software Products must comply with applicable laws and regulations (including trade compliance laws and regulations).
+
+e. You will not, and will not permit, assist or cause any third party to use, modify, copy, reproduce, create derivative works of, or distribute the Software Products (or any derivative works thereof, works incorporating the Software Products, or any data produced by the Software), in whole or in part, for in any manner that infringes, misappropriates, or otherwise violates any third-party rights.
+
+3. DISCLAIMER OF WARRANTY
+
+UNLESS REQUIRED BY APPLICABLE LAW, THE SOFTWARE PRODUCTS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE SOFTWARE PRODUCTS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE SOFTWARE PRODUCTS AND ANY OUTPUT AND RESULTS.
+
+4. LIMITATION OF LIABILITY
+
+IN NO EVENT WILL IDEA OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN IF IDEA OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY OF ANY OF THE FOREGOING.
+
+5. INDEMNIFICATION
+
+You will indemnify, defend and hold harmless IDEA and our subsidiaries and affiliates, and each of our respective shareholders, directors, officers, employees, agents, successors, and assigns (collectively, the “IDEA Parties”) from and against any losses, liabilities, damages, fines, penalties, and expenses (including reasonable attorneys’ fees) incurred by any IDEA Party in connection with any claim, demand, allegation, lawsuit, proceeding, or investigation (collectively, “Claims”) arising out of or related to: (a) your access to or use of the Software Products (as well as any results or data generated from such access or use); (b) your violation of this License; or (c) your violation, misappropriation or infringement of any rights of another (including intellectual property or other proprietary rights and privacy rights). You will promptly notify the IDEA Parties of any such Claims, and cooperate with IDEA Parties in defending such Claims. You will also grant the IDEA Parties sole control of the defense or settlement, at IDEA’s sole option, of any Claims. This indemnity is in addition to, and not in lieu of, any other indemnities or remedies set forth in a written agreement between you and IDEA or the other IDEA Parties.
+
+6. TERMINATION; SURVIVAL
+
+a. This License will automatically terminate upon any breach by you of the terms of this License.
+
+b. If you institute litigation or other proceedings against IDEA or any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Software Products, or any portion of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or claim is filed or instituted.
+
+c. The following sections survive termination of this License: 2 (Redistribution and use), 3 (Disclaimers of Warranty), 4 (Limitation of Liability), 5 (Indemnification), 6 (Termination; Survival), 7 (Trademarks) and 8 (Applicable Law; Dispute Resolution). 
+
+7. TRADEMARKS
+
+Licensee has not been granted any trademark license as part of this License and may not use any name or mark associated with IDEA without the prior written permission of IDEA, except to the extent necessary to make the reference required by the attribution notice of this Agreement.
+
+8. APPLICABLE LAW; DISPUTE RESOLUTION
+
+This License will be governed and construed under the laws of the People’s Republic of China without regard to conflicts of law provisions. The parties expressly agree that the United Nations Convention on Contracts for the International Sale of Goods will not apply. Any suit or proceeding arising out of or relating to this License will be brought in the courts, as applicable, in Shenzhen, Guangdong, and each party irrevocably submits to the jurisdiction and venue of such courts.

humanref/manifest.json

@@ -0,0 +1,66 @@
+{
+  "dataset": "humanref",
+  "source": "https://huggingface.co/datasets/IDEA-Research/HumanRef-CoT-45k",
+  "subset": false,
+  "source_counts": {
+    "humanref_cot": {
+      "source_annotation_rows": 37057,
+      "images": 37057,
+      "positive": 37057,
+      "negative": 0
+    },
+    "humanref_cot_rejection": {
+      "source_annotation_rows": 8351,
+      "images": 8351,
+      "positive": 0,
+      "negative": 8351
+    }
+  },
+  "source_checksums": {
+    "humanref_cot.annotations.tsv": "7e384a8cfa851364bf9d198057d385685aab6a17fa4ef369603e326080400d6a",
+    "humanref_cot.images.tsv": "103a372c54ecf907e198bca8c74e3d59de50dca03d8d30e22537be51f2fcf85a",
+    "humanref_cot_rejection.annotations.tsv": "a3c8da68b60f2f8f4c70c6256ffec5da780b1aa5d926ef102e1eff5ff593c7df",
+    "humanref_cot_rejection.images.tsv": "da5369c6d1d06de0acd5ee1bb004930cc5ce619b5c1f7adf1c7e48b45105e727"
+  },
+  "image_record_policy": "One stored image per referenced source TSV record; identical contents are not deduplicated.",
+  "training_warning": "Raw duplicates, conflicts, and out-of-image boxes are preserved. Source think text is provenance, not a direct-box target.",
+  "format": "rex-toy-tsv",
+  "format_version": 1,
+  "toy_source": "Mountchicken/Rex-Omni-Finetune-ToyData@a8ce2750cb60921cd1b734227bb5e1e2d5ad55ef",
+  "split": "train",
+  "coordinate_format": "absolute pixel xyxy; original pixel xywh preserved in source_annotation",
+  "image_encoding": "base64 of original bytes; no resize, EXIF transpose, or re-encoding",
+  "annotations": 45408,
+  "images": 45408,
+  "positive_rows": 37057,
+  "files": {
+    "train.images.tsv": {
+      "sha256": "78bcb4d6ce3018a43e59f3c98016cfc50225d7cf23a202f3fb48e52689564822",
+      "bytes": 11623877640
+    },
+    "train.annotations.tsv": {
+      "sha256": "aa5d37c81d9802b4953d780c344580e95381449c0f297a9148334990ca1dc8a6",
+      "bytes": 133314604
+    },
+    "train.annotations.tsv.lineidx": {
+      "sha256": "38a17bfd8c37bdae60c806c7577572ce151ba81d9d19daf399e142aaa0b04de0",
+      "bytes": 416353
+    },
+    "train.images.tsv.lineidx": {
+      "sha256": "bd8b543dd2f556e3a4d47268b8da188ae77394e745eb36acad4aaa2a1fc871f9",
+      "bytes": 501624
+    },
+    "train.positive.annotations.tsv.lineidx": {
+      "sha256": "7e2fd10d64f8ca98b3d8cf53c28318bbbc0a42c505253d9a330efd1a10860819",
+      "bytes": 332843
+    },
+    "images.jsonl": {
+      "sha256": "aaa30a5a437a788c542bb843d8d21a57816733a5eb5bce0b947af9c4d2387763",
+      "bytes": 7882352
+    }
+  },
+  "selection": "all source train records, original order; no deduplication or leakage filtering",
+  "ready_for_training": false,
+  "source_revision": "9e1afd27358d774f626a78992c7b716fa403da54",
+  "source_checksums_match_prior_audit": true
+}

humanref/train.annotations.tsv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa5d37c81d9802b4953d780c344580e95381449c0f297a9148334990ca1dc8a6
+size 133314604

humanref/train.images.tsv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78bcb4d6ce3018a43e59f3c98016cfc50225d7cf23a202f3fb48e52689564822
+size 11623877640

humanref/validation.json

@@ -0,0 +1,14 @@
+{
+  "status": "passed",
+  "dataset": "humanref",
+  "images": 45408,
+  "annotations": 45408,
+  "positive_rows": 37057,
+  "negative_rows": 8351,
+  "target_boxes": 86937,
+  "full_pixel_decode": false,
+  "original_image_bytes_verified": true,
+  "all_byte_offsets_verified": true,
+  "raw_annotations_verified": true,
+  "subset": false
+}

loader.py

@@ -0,0 +1,42 @@
+"""Read raw pixel annotations; no coordinate quantization or target formatting."""
+import base64
+import io
+import json
+from pathlib import Path
+from PIL import Image
+
+
+class RawTSVDataset:
+    def __init__(self, root, positive_only=False):
+        self.root = Path(root)
+        self.manifest = json.loads((self.root/'manifest.json').read_text())
+        name = 'train.positive.annotations.tsv.lineidx' if positive_only else 'train.annotations.tsv.lineidx'
+        self.offsets = [int(x) for x in (self.root/name).read_text().splitlines()]
+
+    def __len__(self):
+        return len(self.offsets)
+
+    def __getitem__(self, index):
+        # Open per request: safe under fork/spawn and concurrent readers.
+        with (self.root/'train.annotations.tsv').open('rb') as f:
+            f.seek(self.offsets[index])
+            offset, payload = f.readline().split(b'\t', 1)
+        annotation = json.loads(payload)
+        with (self.root/'train.images.tsv').open('rb') as f:
+            f.seek(int(offset))
+            key, encoded = f.readline().rstrip(b'\n').split(b'\t', 1)
+        if int(key) != int(offset):
+            raise ValueError('Image offset mismatch')
+        data = base64.b64decode(encoded, validate=True)
+        return {'image_bytes': data, 'annotation': annotation}
+
+    @staticmethod
+    def decode(sample):
+        # Do not EXIF-transpose: all coordinates describe the stored raster.
+        with Image.open(io.BytesIO(sample['image_bytes'])) as source:
+            image = source.convert('RGB')
+        image.info.pop('exif', None)
+        ann = sample['annotation']
+        if image.size != (ann['width'], ann['height']):
+            raise ValueError('Image size mismatch')
+        return {'image': image, 'annotation': ann}

requirements.txt

@@ -0,0 +1 @@
+Pillow>=10,<13