Upload README.md with huggingface_hub

README.md

@@ -1,77 +1,13 @@
 ---
-dataset_info:
-- config_name: films
-  features:
-  - name: id
-    dtype: string
-  - name: image
-    dtype: image
-  - name: choices
-    list: string
-  - name: answer
-    dtype: string
-  - name: answer_index
-    dtype: int32
-  - name: answer_title
-    dtype: string
-  - name: title
-    dtype: string
-  - name: artist
-    dtype: string
-  - name: year
-    dtype: int32
-  - name: genre
-    dtype: string
-  - name: average_color
-    dtype: string
-  - name: colorfulness
-    dtype: float32
-  - name: seconds
-    dtype: int32
-  - name: source_url
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 406214
-    num_examples: 100
-  download_size: 421047
-  dataset_size: 406214
-- config_name: youtube
-  features:
-  - name: id
-    dtype: string
-  - name: image
-    dtype: image
-  - name: choices
-    list: string
-  - name: answer
-    dtype: string
-  - name: answer_index
-    dtype: int32
-  - name: answer_title
-    dtype: string
-  - name: title
-    dtype: string
-  - name: artist
-    dtype: string
-  - name: year
-    dtype: int32
-  - name: genre
-    dtype: string
-  - name: average_color
-    dtype: string
-  - name: colorfulness
-    dtype: float32
-  - name: seconds
-    dtype: int32
-  - name: source_url
-    dtype: string
-  splits:
-  - name: test
-    num_bytes: 278435
-    num_examples: 100
-  download_size: 289340
-  dataset_size: 278435
+license: mit
+task_categories:
+- visual-question-answering
+- multiple-choice
+language:
+- en
+pretty_name: MovieColorBench
+size_categories:
+- n<1K
 configs:
 - config_name: films
   data_files:
@@ -82,3 +18,46 @@ configs:
   - split: test
     path: youtube/test-*
 ---
+
+# MovieColorBench
+
+Can a model recognize a famous video from just its colors?
+
+Each image is a barcode of one video. Every stripe is the average color of a few seconds, in order from start to end. The model picks which video it is out of 10 options, so guessing gets 10%.
+
+| config | questions | what's in it |
+|--------|-----------|--------------|
+| `films` | 100 | feature films |
+| `youtube` | 100 | the most viewed YouTube videos, mostly music videos |
+
+```python
+from datasets import load_dataset
+ds = load_dataset("loganbolton/MovieColorBench", "films", split="test")
+ds[0]["image"], ds[0]["choices"], ds[0]["answer"]
+```
+
+## Fields
+
+| field | meaning |
+|-------|---------|
+| `image` | the barcode, 1000x200 PNG |
+| `choices` | the 10 options |
+| `answer`, `answer_index`, `answer_title` | the right option as a letter (A to J), an index (0 to 9) and text |
+| `title`, `artist`, `year`, `genre` | about the video |
+| `average_color` | mean color of the whole video as hex |
+| `colorfulness` | how saturated the video is, near 0 for black and white |
+| `seconds` | runtime |
+| `source_url` | where the video came from |
+
+The wrong options are picked to be plausible. Same genre, close in year, similar colorfulness.
+
+## Running a model
+
+The code is at https://github.com/LoganBolton/MovieColorBench. Any OpenRouter model runs with one command.
+
+```bash
+export OPENROUTER_API_KEY=sk-or-...
+uv run run.py --model openai/gpt-6.1-sol
+```
+
+Only barcodes are shared here, no frames from the videos.