Upload README.md with huggingface_hub

README.md

@@ -1,73 +1,84 @@
 ---
 dataset_info:
   features:
-  - name: id
-    dtype: string
-  - name: split
-    dtype: string
-  - name: image
-    dtype: image
-  - name: image_path
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: text_only_question
-    dtype: string
-  - name: trivial_text_only_question
-    dtype: string
-  - name: hour
-    dtype: int32
-  - name: minute
-    dtype: int32
-  - name: second
-    dtype: int32
-  - name: show_second_hand
-    dtype: bool
-  - name: answer_format
-    dtype: string
-  - name: style_name
-    dtype: string
-  - name: theme
-    dtype: string
-  - name: resolution
-    list: int32
-  - name: width
-    dtype: int32
-  - name: height
-    dtype: int32
-  - name: layout
-    dtype: string
-  - name: dial_shape
-    dtype: string
-  - name: hour_deg
-    dtype: float32
-  - name: minute_deg
-    dtype: float32
-  - name: second_deg
-    dtype: float32
-  - name: metadata_json
-    dtype: string
+    - name: image
+      dtype: image
+    - name: question
+      dtype: string
+    - name: answer
+      dtype: string
+    - name: text_only_question
+      dtype: string
+    - name: trivial_text_only_question
+      dtype: string
   splits:
-  - name: train
-    num_bytes: 719703991
-    num_examples: 5000
-  - name: dev
-    num_bytes: 45182270
-    num_examples: 300
-  - name: test
-    num_bytes: 75768209
-    num_examples: 500
-  download_size: 834158855
-  dataset_size: 840654470
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: dev
-    path: data/dev-*
-  - split: test
-    path: data/test-*
+    - name: train
+      num_examples: 5000
+    - name: dev
+      num_examples: 300
+    - name: test
+      num_examples: 500
+task_categories:
+  - visual-question-answering
+  - image-to-text
+language:
+  - en
+pretty_name: MeasureBenchClock-Diverse-Synthetic
 ---
+
+# MeasureBenchClock-Diverse-Synthetic
+
+MeasureBenchClock-Diverse-Synthetic is a synthetic analog clock-reading dataset generated using a modified and diversified version of the MeasureBench clock-generation code.
+
+This dataset is intended for supervised finetuning and reinforcement-learning experiments on analog-clock reading. It is not an official MeasureBench release.
+
+## Generation
+
+Images are generated using a configurable synthetic analog clock renderer. The generator varies resolution, clock placement, dial shape, color palette, background, hand styles, tick marks, numerals, second-hand visibility, and clock layout.
+
+## Answer policy
+
+If the second hand is visible, the answer is given in 12-hour `H:MM:SS` format.
+
+If the second hand is not visible, the answer is given as the nearest minute in 12-hour `H:MM` format. Hidden seconds may still affect the rendered hour and minute hand positions, but the model is not asked to predict hidden seconds.
+
+No interval labels are used.
+
+## Fields
+
+- `id`: Unique sample ID.
+- `split`: Dataset split.
+- `image`: Generated clock image.
+- `image_path`: Relative path to the generated PNG.
+- `question`: Image-based clock-reading prompt.
+- `answer`: Gold answer string.
+- `text_only_question`: Angle-based text-only version of the question.
+- `trivial_text_only_question`: Tick-position text-only version of the question.
+- `hour`, `minute`, `second`: Underlying rendered time.
+- `show_second_hand`: Whether the second hand is visible.
+- `answer_format`: Either `H:MM:SS` or `H:MM`.
+- `style_name`: High-level generator style.
+- `theme`: Sampled visual theme.
+- `layout`: Clock placement layout.
+- `dial_shape`: Dial shape.
+- `width`, `height`: Image resolution.
+- `hour_deg`, `minute_deg`, `second_deg`: Hand angles in degrees clockwise from 12 o'clock.
+- `metadata_json`: Additional generator metadata serialized as JSON.
+
+## Intended use
+
+This dataset is intended for training and evaluating models on analog-clock reading, especially for experiments involving synthetic data, supervised finetuning, multimodal reasoning, and follow-up RL training.
+
+## Limitations
+
+The images are synthetic and may not fully match real-world clock photos. Some visual styles are deliberately diverse to improve robustness, but the dataset should not be treated as a substitute for real clock images.
+
+## Showcase
+
+The following showcase contains 50 generated clocks sampled from the configured YAML generation settings.
+
+![50 generated analog clock samples](generated_data/clock_showcase.png)
+
+The following smaller showcase contains 4 generated clocks with the corresponding text-only questions displayed under each image.
+
+![Generated analog clock samples with text-only questions](generated_data/clock_showcase_with_questions.png)