Clean dataset card: 39 per-benchmark configs only

README.md

@@ -18,12 +18,8 @@ tags:
   - lottie
 pretty_name: "GDB: GraphicDesignBench"
 size_categories:
-  - 1K<n<10K
+  - 10K<n<100K
 configs:
-  - config_name: all
-    data_files:
-      - split: train
-        path: "all/train-*"
   - config_name: category-1
     data_files:
       - split: train
@@ -180,34 +176,7 @@ configs:
     data_files:
       - split: train
         path: "typography-8/train-*"
-  - config_name: category
-    data_files:
-      - split: train
-        path: "category/train-*"
-  - config_name: layout
-    data_files:
-      - split: train
-        path: "layout/train-*"
-  - config_name: lottie
-    data_files:
-      - split: train
-        path: "lottie/train-*"
-  - config_name: svg
-    data_files:
-      - split: train
-        path: "svg/train-*"
-  - config_name: template
-    data_files:
-      - split: train
-        path: "template/train-*"
-  - config_name: temporal
-    data_files:
-      - split: train
-        path: "temporal/train-*"
-  - config_name: typography
-    data_files:
-      - split: train
-        path: "typography/train-*"
+default_config_name: svg-1
 ---
 
 # GDB: GraphicDesignBench
@@ -215,19 +184,36 @@ configs:
 **A Real-World Benchmark for Graphic Design AI**
 
 GDB is the first large-scale benchmark grounded in the work marketing and branding teams ship every day.
-It provides a reproducible signal for where AI models stand on professional design tasks, and what still
-needs to change before they can do real design work.
+It covers 39 tasks across 7 design domains, testing both understanding (e.g. classify a font, detect components)
+and generation (e.g. produce an SVG, render styled text, generate a layout).
 
 - **Paper**: [arXiv:2604.04192](https://arxiv.org/abs/2604.04192)
 - **Code**: [github.com/purvanshi/lica-bench](https://github.com/purvanshi/lica-bench)
 - **Blog**: [lica.world/blog/gdb-real-world-benchmark-for-graphic-design](https://lica.world/blog/gdb-real-world-benchmark-for-graphic-design)
-- **Results**: [Explore all task results](https://lica.world/blog/gdb-real-world-benchmark-for-graphic-design)
 
-## Benchmark Structure
+## Quick Start
 
-39 benchmarks across 7 design domains and 2 task modes (understanding + generation):
+```python
+from datasets import load_dataset
+
+# Load one benchmark
+ds = load_dataset("lica-world/GDB", "svg-1", split="train")
+
+# Load several benchmarks together
+from datasets import concatenate_datasets
+configs = ["svg-1", "svg-2", "svg-3"]
+ds = concatenate_datasets([
+    load_dataset("lica-world/GDB", c, split="train") for c in configs
+])
+```
+
+## Dataset Structure
+
+Each benchmark is a separate config. 32764 samples total across 39 benchmarks.
+
+### Overview
 
-| Domain | Understanding | Generation | Total |
+| Domain | Understanding | Generation | Total Tasks |
 |--------|:---:|:---:|:---:|
 | Layout | 4 | 4 | 8 |
 | Typography | 6 | 2 | 8 |
@@ -238,42 +224,71 @@ needs to change before they can do real design work.
 | Lottie | 0 | 2 | 2 |
 | **Total** | **23** | **16** | **39** |
 
-## Usage
+### All Benchmarks
 
-```python
-from datasets import load_dataset
-
-# Load a specific benchmark
-ds = load_dataset("lica-world/GDB", "svg-1")
-
-# Load all benchmarks
-ds = load_dataset("lica-world/GDB", "all")
-
-# Load all benchmarks in a domain
-ds = load_dataset("lica-world/GDB", "typography")
-```
+| Config | Task | Domain | Type | Samples |
+|--------|------|--------|------|--------:|
+| `category-1` | Fine-Grained Category Classification | category | understanding | 989 |
+| `category-2` | User Intent Prediction | category | understanding | 989 |
+| `layout-1` | Intent-to-Layout Generation | layout | generation | 100 |
+| `layout-2` | Visual-Element-Contents-Aware Layout Generation | layout | generation | 1022 |
+| `layout-3` | Aspect-Ratio Adaptation | layout | generation | 13 |
+| `layout-4` | Aspect Ratio Classification | layout | understanding | 989 |
+| `layout-5` | Component Count | layout | understanding | 989 |
+| `layout-6` | Component Classification | layout | understanding | 2943 |
+| `layout-7` | Component Detection | layout | understanding | 988 |
+| `layout-8` | Layer-Aware Object Insertion | layout | generation | 100 |
+| `lottie-1` | Text-to-Lottie Generation | lottie | generation | 50 |
+| `lottie-2` | Image-Text-to-Lottie Generation | lottie | generation | 50 |
+| `svg-1` | SVG Perceptual Q/A | svg | understanding | 300 |
+| `svg-2` | SVG Semantic Q/A | svg | understanding | 300 |
+| `svg-3` | SVG Bug Fixing | svg | understanding | 900 |
+| `svg-4` | SVG Code Optimization | svg | understanding | 300 |
+| `svg-5` | SVG Style Editing | svg | understanding | 900 |
+| `svg-6` | Text-to-SVG Generation | svg | generation | 300 |
+| `svg-7` | Image-to-SVG Generation | svg | generation | 300 |
+| `svg-8` | Image-Text-to-SVG Generation | svg | generation | 300 |
+| `template-1` | Pairwise Layout Matching | template | understanding | 1000 |
+| `template-2` | Layout Retrieval | template | understanding | 200 |
+| `template-3` | Layout Clustering | template | understanding | 50 |
+| `template-4` | Style Completion | template | generation | 140 |
+| `template-5` | Color Transfer | template | generation | 200 |
+| `temporal-1` | Keyframe Ordering | temporal | understanding | 100 |
+| `temporal-2` | Motion Type Classification | temporal | understanding | 1896 |
+| `temporal-3` | Animation Property Extraction | temporal | understanding | 848 |
+| `temporal-4` | Animation Parameter Generation | temporal | generation | 10 |
+| `temporal-5` | Motion Trajectory Generation | temporal | generation | 10 |
+| `temporal-6` | Short-Form Video Layout Generation | temporal | generation | 10 |
+| `typography-1` | Font Family Classification | typography | understanding | 2556 |
+| `typography-2` | Text Color Estimation | typography | understanding | 2498 |
+| `typography-3` | Text Params Estimation | typography | understanding | 2556 |
+| `typography-4` | Style Ranges | typography | understanding | 2556 |
+| `typography-5` | Curved Text | typography | understanding | 2556 |
+| `typography-6` | Text Rotation | typography | understanding | 2556 |
+| `typography-7` | Styled Text Rendering to Layout | typography | generation | 100 |
+| `typography-8` | Styled Text Element Generation | typography | generation | 100 |
 
 ## Schema
 
-Each sample contains:
+Every sample has a consistent schema across all configs:
 
 | Field | Type | Description |
 |-------|------|-------------|
-| `sample_id` | string | Unique sample identifier |
-| `benchmark_id` | string | Benchmark ID (e.g., `svg-1`, `typography-3`) |
-| `domain` | string | Domain: layout, typography, svg, template, temporal, category, lottie |
+| `sample_id` | string | Unique identifier |
+| `benchmark_id` | string | Config name (e.g. `svg-1`) |
+| `domain` | string | One of: layout, typography, svg, template, temporal, category, lottie |
 | `task_type` | string | `understanding` or `generation` |
-| `benchmark_name` | string | Human-readable benchmark name |
-| `prompt` | string | The evaluation prompt sent to the model |
-| `ground_truth` | string | Expected answer (JSON-serialized for complex types) |
-| `image` | Image | Input image when applicable |
-| `media_path` | string | Relative path to media file (video, additional images) |
+| `benchmark_name` | string | Human-readable task name |
+| `prompt` | string | The evaluation prompt |
+| `ground_truth` | string | Expected answer (JSON-serialized for structured types) |
+| `image` | Image | Input image when applicable, `null` otherwise |
+| `media_path` | string | Relative path to media (for video or additional images) |
 | `media_type` | string | `image`, `video`, or `none` |
-| `metadata` | string | JSON with task-specific extra fields (SVG code, options, etc.) |
+| `metadata` | string | JSON with task-specific extra fields (SVG code, options, style specs, etc.) |
 
 ## Evaluation
 
-Install `lica-bench` and use the built-in evaluation pipeline:
+Use the `lica-bench` package for scoring:
 
 ```bash
 pip install lica-bench
@@ -284,11 +299,12 @@ from design_benchmarks.registry import BenchmarkRegistry
 
 registry = BenchmarkRegistry()
 registry.discover()
+
 bench = registry.get("svg-1")
-scores = bench.evaluate(predictions, ground_truth)
+# scores = bench.evaluate(predictions, ground_truths)
 ```
 
-See the [lica-bench README](https://github.com/purvanshi/lica-bench) for full usage.
+See the [lica-bench README](https://github.com/purvanshi/lica-bench) for full CLI and API usage.
 
 ## Citation