Add dataset card

README.md

@@ -1,10 +1,29 @@
 ---
+language:
+- en
+license: cc-by-sa-4.0
+task_categories:
+- image-to-text
+- text-to-image
+- image-classification
+- zero-shot-image-classification
+tags:
+- patent
+- design-patent
+- technical-drawing
+- wireframe
+- multimodal
+- retrieval
+- reconstruction
+- USPTO
 dataset_info:
   features:
   - name: figure_id
     dtype: string
   - name: patent_id
     dtype: string
+  - name: image
+    dtype: image
   - name: patent_title
     dtype: string
   - name: caption
@@ -23,37 +42,94 @@ dataset_info:
     dtype: int32
   - name: n_figures_in_patent
     dtype: int32
-  - name: year
-    dtype: int32
-  - name: svg
-    dtype: string
   - name: sibling_figure_ids
-    list: string
+    sequence: string
+  - name: reference_numerals
+    sequence:
+      struct:
+      - name: numeral
+        dtype: string
+      - name: label
+        dtype: string
   - name: locarno_class
     dtype: string
   - name: patent_date
     dtype: string
-  - name: image
-    dtype: image
+  - name: year
+    dtype: int32
   splits:
   - name: train
-    num_bytes: 479249
     num_examples: 397
   - name: validation
-    num_bytes: 38808
     num_examples: 39
   - name: test
-    num_bytes: 103733
     num_examples: 64
-  download_size: 119260
-  dataset_size: 621790
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
-  - split: test
-    path: data/test-*
 ---
+
+# Patent Wireframes
+
+Structured patent figure dataset for multimodal understanding — generation, reconstruction, retrieval, and cross-modal association.
+
+## Overview
+
+Each record is a single patent figure paired with:
+- Full patent text (drawing description, detailed description, brief summary, claims)
+- IMPACT-generated figure caption
+- Extracted reference numeral mappings (numeral → component label)
+- Sibling figure links (other views of the same object)
+- Classification metadata (Locarno codes, dates)
+
+## Sources
+
+- **Images + captions**: [IMPACT](https://huggingface.co/datasets/AI4Patents/IMPACT) (CC-BY-SA-4.0)
+- **Patent text**: [PatentsView](https://patentsview.org/) (CC-BY-4.0)
+- **Scope**: US design patents, initially 2022
+
+## Schema
+
+| Field | Type | Description |
+|---|---|---|
+| `figure_id` | string | Unique: `{patent_id}_{figure_number}` |
+| `patent_id` | string | USPTO design patent number |
+| `image` | image | Figure raster |
+| `patent_title` | string | Patent title |
+| `caption` | string | IMPACT-generated caption |
+| `drawing_description` | string | Drawing description from patent text |
+| `detailed_description` | string | Full detailed description |
+| `brief_summary` | string | Brief summary |
+| `claims` | string | Patent claims |
+| `viewpoint` | string | front / side / perspective / etc. |
+| `figure_number` | int32 | Figure index within patent |
+| `n_figures_in_patent` | int32 | Total figures in patent |
+| `sibling_figure_ids` | list[string] | Other figure IDs from same patent |
+| `reference_numerals` | list[{numeral, label}] | Extracted component mappings |
+| `locarno_class` | string | Locarno classification code |
+| `patent_date` | string | Grant date |
+| `year` | int32 | Grant year |
+
+## Intended uses
+
+- **Reconstruction**: predict figure from text description (or vice versa)
+- **Retrieval**: text-to-figure, figure-to-figure, component-to-figure
+- **Generation**: produce patent figures from structured descriptions
+- **Multi-view reasoning**: predict unseen views given other views
+- **Component understanding**: reference numeral grounding
+
+## Splits
+
+Train/val/test split at the **patent level** (not figure level) to prevent
+multi-view data leakage. 80/10/10 ratio, random with seed 42.
+
+## Citation
+
+If you use this dataset, please cite:
+
+```bibtex
+@dataset{patent_wireframes,
+  title={Patent Wireframes: A Structured Dataset for Multimodal Patent Figure Understanding},
+  year={2026},
+  url={https://huggingface.co/datasets/midah/patent-wireframes},
+}
+```
+
+Built on [IMPACT](https://openreview.net/forum?id=l0Ydsl10ci) (NeurIPS 2024) and [PatentsView](https://patentsview.org/).