Separate canonical loading from image preview

README.md

@@ -9,6 +9,38 @@ task_categories:
 - image-to-image
 license: other
 dataset_info:
+- config_name: default
+  features:
+  - name: id
+    dtype: string
+  - name: capability
+    dtype: string
+  - name: task
+    dtype: string
+  - name: image_path
+    dtype: string
+  - name: media_paths
+    sequence: string
+  - name: mask_path
+    dtype: string
+  - name: question
+    dtype: string
+  - name: answer
+    dtype: string
+  - name: answer_type
+    dtype: string
+  - name: choices
+    sequence: string
+  - name: source
+    dtype: string
+  - name: source_id
+    dtype: string
+  - name: metadata_json
+    dtype: string
+  splits:
+  - name: test
+    num_examples: 470
+- config_name: preview
   features:
   - name: image
     dtype: image
@@ -29,6 +61,10 @@ dataset_info:
     num_examples: 470
 configs:
 - config_name: default
+  data_files:
+  - split: test
+    path: data/spatialgen_bench_full.parquet
+- config_name: preview
   data_files:
   - split: test
     path: data/spatialgen_bench_viewer.parquet
@@ -51,6 +87,7 @@ SpatialGen-Bench is a diagnostic benchmark for spatial cognition in image-genera
 SpatialGen-Bench/
 ├── data/
 │   ├── spatialgen_bench.jsonl
+│   ├── spatialgen_bench_full.parquet
 │   └── spatialgen_bench_viewer.parquet
 ├── benchmarks/
 │   ├── perception/
@@ -74,19 +111,21 @@ Each JSONL record contains:
 - `question`, string-normalized `answer`, `answer_type`, and `choices`
 - `source`, `source_id`, and serialized task-specific `metadata_json`
 
-The default Hugging Face configuration is a concise browsing view backed by
-`spatialgen_bench_viewer.parquet`. It adds compact 512px previews and keeps only
-the fields useful for inspection. The canonical JSONL retains the complete schema
-and original order for evaluation.
+The default Hugging Face configuration mirrors the complete canonical JSONL schema
+and original evaluation order. The optional `preview` configuration adds compact
+512px images, task-grouped rows, and only the fields useful for inspection.
 
-Load the browsing view with Hugging Face Datasets:
+Load either view with Hugging Face Datasets:
 
 ```python
 from datasets import load_dataset
 
-preview = load_dataset("wx91726/SpatialGen-Bench", split="test")
+full = load_dataset("wx91726/SpatialGen-Bench", split="test")
+preview = load_dataset("wx91726/SpatialGen-Bench", "preview", split="test")
 ```
 
+Browse the image-rich [preview configuration](https://huggingface.co/datasets/wx91726/SpatialGen-Bench/viewer/preview/test).
+
 Media fields contain paths relative to the repository's `benchmarks/` directory. Download the complete evaluation package with:
 
 ```bash

data/spatialgen_bench_full.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99cb447bcba5b487d9b9f0a533450e12c2a21b951795501bcf7c711a6db80e04
+size 124762