Upload README.md with huggingface_hub

README.md

@@ -1,5 +1,5 @@
 ---
-license: mit
+license: cc-by-nc-4.0
 task_categories:
   - object-detection
 language:
@@ -8,60 +8,108 @@ tags:
   - document-layout
   - khmer
   - coco
-pretty_name: ARDB Daily Bulletin Layout (COCO) v1
+  - object-detection
+pretty_name: ARDB Daily Bulletin Layout (COCO)
+size_categories:
+  - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*
+      - split: validation
+        path: data/validation-*
+      - split: test
+        path: data/test-*
 ---
 
-# ARDB Daily Bulletin Layout Detection — v1 (COCO format)
+# ARDB Daily Bulletin Layout Detection (COCO)
+
+A document-layout **object-detection** dataset built from **ARDB (Agricultural and Rural
+Development Bank of Cambodia) daily market-price bulletins** — born-digital Khmer-language PDFs,
+one price table per page. Every page is annotated with bounding boxes for five layout regions
+(table, text, section header, page furniture, picture).
+
+Each row is a full page image rendered at **200 DPI (2000 × 2000 px, JPG)** with its box
+annotations embedded, so the Dataset Viewer shows the image beside its labels.
+
+## Dataset structure
+
+| Split      | Pages | Boxes | Source documents |
+|------------|------:|------:|-----------------:|
+| train      |    73 |   341 | 24 bulletins     |
+| validation |     9 |    42 |  3 bulletins     |
+| test       |     9 |    42 |  3 bulletins     |
+| **total**  | **91**| **425**| **30 bulletins** |
+
+### Fields
+
+| Column      | Type            | Description |
+|-------------|-----------------|-------------|
+| `image`     | image           | Embedded page image (JPG, 2000 × 2000). |
+| `image_id`  | int64           | Row index within the split. |
+| `file_name` | string          | Source image file name. |
+| `doc_id`    | string          | Identifier of the originating bulletin. |
+| `source`    | string          | Originating PDF (provenance). |
+| `width`     | int64           | Image width in pixels. |
+| `height`    | int64           | Image height in pixels. |
+| `objects`   | struct of lists | Per-box annotations (see below). |
 
-Document-layout detection dataset built from **ARDB (Agricultural and Rural Development Bank
-of Cambodia) daily market-price bulletins** — born-digital Khmer PDFs with one price table per
-page. Purpose: fine-tune a layout detector (YOLO-class) for a Khmer financial-document OCR
-pipeline at GDDE/MEF.
+`objects` is a struct of parallel lists carrying the full COCO annotation fields:
 
-## Contents
+- `id` — annotation id
+- `bbox` — `[x, y, width, height]` in pixels (COCO convention)
+- `category_id` — class index (0–4, see below)
+- `category` — human-readable class name
+- `area` — box area in px²
+- `iscrowd` — always `0`
+- `score` — annotation confidence (`1.0`, human-verified)
 
-| Split | Pages | Boxes | Documents |
-|---|---|---|---|
-| train | 73 | 341 | 24 bulletins |
-| valid | 9 | 42 | 3 bulletins |
-| test | 9 | 42 | 3 bulletins |
+### Classes
 
-The dataset is stored as self-contained Parquet shards under `data/` (one per split): page
-images (JPG, rendered at 200 DPI) are embedded, so the Data Viewer shows one tabular row per
-page — the image beside its annotations.
+| id | name             | Boxes | Notes                                   |
+|----|------------------|------:|-----------------------------------------|
+| 0  | `Table`          |    91 | The price table — exactly one per page, covering the full table including header row and label columns. |
+| 1  | `Text`           |    30 | Body / paragraph text.                  |
+| 2  | `Section-Header` |    31 | Headings and titles.                    |
+| 3  | `Page-Furniture` |   182 | Page headers and footers.               |
+| 4  | `Picture`        |    91 | Logos and stamps.                       |
+
+## Usage
 
 ```python
 from datasets import load_dataset
-ds = load_dataset("<repo_id>")   # splits: train / validation / test
+
+ds = load_dataset("Soxavin/ardb-layout-coco-v2")   # splits: train / validation / test
+row = ds["train"][0]
+row["image"]          # PIL image
+row["objects"]["bbox"]        # list of [x, y, w, h]
+row["objects"]["category"]    # list of class names
 ```
 
-**Columns:** `image` (embedded page image) · `image_id` · `file_name` · `doc_id` ·
-`source` (originating bulletin PDF — provenance) · `width` · `height` · `objects`, a struct
-of per-box lists carrying the full COCO fields: `id`, `bbox` (`[x, y, width, height]` in
-pixels), `category_id`, `category` (readable label), `area`, `iscrowd`, `score`
-(pseudo-label confidence; 1.0 after human correction).
-
-**Classes (5), category id order:** 0 `Table`, 1 `Text`, 2 `Section-Header`,
-3 `Page-Furniture` (page headers/footers), 4 `Picture` (logos/stamps).
-Exactly one `Table` box per page, covering the full table **including header row and label
-columns** (a deliberate rule: off-the-shelf detectors clipping label columns was a measured
-failure mode).
-
-## Provenance & method
-
-1. Source PDFs: publicly published ARDB daily bulletins (June–July 2026), rendered at 200 DPI.
-2. **Pseudo-labels** generated by Surya layout detection (`surya-ocr` 2.x), mapped to the
-   5-class set, confidence-filtered at 0.5.
-3. **Human correction pass** in Roboflow (model-assisted labeling): fragmented table boxes
-   merged to one per page, box edges tightened, labels fixed, junk boxes removed.
-4. Split is **by document** (no bulletin's pages straddle splits) to prevent
-   near-duplicate-page leakage; assignment is deterministic (seed 0).
-
-## Caveats
-
-- **Single-template corpus:** all 30 bulletins share one layout. valid/test measure
-  label quality, not cross-template generalization — evaluate generalization on held-out
-  document types (budget/TOFE reports, gas-price bulletins) via the downstream pipeline harness.
-- Pseudo-labels originate from Surya (weights under a modified OpenRAIL-M licence); this
-  dataset is intended for **private/internal training use** pending a licence review for any
-  public redistribution.
+## Data collection & annotation
+
+1. **Source.** Publicly published ARDB daily market-price bulletins, rendered page-by-page at
+   200 DPI.
+2. **Pre-annotation.** Candidate boxes were generated automatically with a document-layout
+   detector and confidence-filtered, then mapped to the five-class set above.
+3. **Human correction.** Every page was reviewed and corrected in Roboflow: fragmented table
+   regions merged to one box per page, box edges tightened, mislabels fixed, and spurious boxes
+   removed. All final annotations carry `score = 1.0`.
+4. **Splitting.** Splits are assigned **by document** (no bulletin's pages appear in more than
+   one split) to prevent near-duplicate-page leakage. Assignment is deterministic.
+
+## Limitations
+
+- **Single template.** All bulletins share one recurring page layout, so the validation and test
+  splits measure annotation quality rather than generalization across document designs.
+- **Small scale.** 91 pages / 30 documents; suitable for fine-tuning and evaluation on this
+  document family, not as a general-purpose layout corpus.
+
+## License
+
+The **annotations** in this dataset (bounding boxes, class labels, and metadata) are released
+under **CC BY-NC 4.0** — free to use for non-commercial research with attribution. The **page
+images** are reproductions of bulletins published by ARDB; they are included for research use
+only, and users are responsible for complying with the source documents' terms. This dataset is
+not affiliated with or endorsed by ARDB.