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%.
Dataset Card
By Logan Bolton, published under mit, revision bd3e89a29c39.
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 |
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.
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.
Structure
films 100 rows
| Split | Rows | Size |
|---|---|---|
| test | 100 | 406.5 KB |
youtube 100 rows
| Split | Rows | Size |
|---|---|---|
| test | 100 | 279.0 KB |
Details
- Repository
- loganbolton/MovieColorBench
- Publisher
- Logan Bolton
- Task category
- Visual question answering
- Tags
- Not stated by the source
- Size category
- n<1K
- Languages
- en
- Revision
- bd3e89a29c392da33dbb3a2eff4a7e3e350a7be5
- Last updated
- 2026-10-02
Files
4 files, 714.7 KB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| films/test-00000-of-00001.parquet | Data | 421.0 KB | ff750f2949ce |
| youtube/test-00000-of-00001.parquet | Data | 289.3 KB | 4a91ce58343c |
| README.md | Documentation | 1.8 KB | — |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- mit
- Access
- No access gate
Released by Logan Bolton through its official repository on Hugging Face. Read the license.