AINovice2005/pixel-art-bench-v1

---
dataset_info:
features:
- name: id
dtype: int64
- name: model_slug
dtype: string
- name: model_name
dtype: string
- name: example_id
dtype: int64
- name: example_name
dtype: string
- name: palette
list: string
- name: grid
list: string
- name: is_appropriate
dtype: bool
- name: height
dtype: int64
- name: width
dtype: int64
- name: num_colors
dtype: int64
- name: input_tokens
dtype: int64
- name: output_tokens
dtype: int64
- name: total_tokens
dtype: int64
- name: cost
dtype: float64
- name: generation_time
dtype: float64
splits:
- name: train
num_bytes: 2496351
num_examples: 1675
download_size: 2298137
dataset_size: 2496351
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: apache-2.0
task_categories:
- text-generation
language:
- en
tags:
- art
- pixel-art
size_categories:
- 1K<n<10K
---

### Pixel Art Benchmark Dataset (Source)

The Pixel Art Benchmark Dataset is a structured collection of pixel-art outputs generated by large language models (LLMs). Each sample consists of a discrete color palette and a grid-based representation of pixel art, along with generation metadata such as token usage, cost, and model provenance.

Each row in the dataset represents a single generated pixel-art sample.

## Encoding Details
Each string in grid represents one row of pixels. Each character corresponds to an index in palette

Example:
{
"palette": ["000000", "ff0000", "ffffff"],
"grid": [
"0110",
"1221",
"0110"
]
}

### Data Source

The dataset is derived from raw generation logs collected via model inference APIs (e.g., OpenRouter-compatible providers). Each record includes full response metadata, from which structured pixel-art outputs are extracted.

### Example Usage

```python
from datasets import load_dataset

ds = load_dataset("AINovice2005/pixel-art-bench-v1")

sample = ds["train"][0]

palette = sample["palette"]
grid = sample["grid"]

# Convert to numeric grid
import numpy as np
img = np.array([[int(c) for c in row] for row in grid])
```