AINovice2005/pixel-art-bench-lite

🎨 Pixel Art Bench Lite

Pixel Art Bench Lite is a structured-output benchmark designed to evaluate small language models on their ability to generate valid, interpretable, and semantically meaningful JSON outputs under strict constraints.

The benchmark is based on Pixel Art Bench focuses on pixel art generation over a fixed 24×24 grid, requiring models to produce outputs that are syntactically correct but also visually coherent.

While many benchmarks evaluate free-form text generation, real-world applications increasingly require:

Strict JSON compliance Structured reasoning Format adherence under constraints

Pixel Art Bench targets this gap by evaluating:

Whether a model can follow rigid output schemas Whether outputs are machine-interpretable Whether generated structures are usable downstream (e.g., rendering)

🧪 Task Definition

Each sample consists of:

A prompt describing a subject A requirement to generate a 24×24 pixel grid

Input

A natural language instruction:

Draw pixel art on a 24x24 grid. Return a JSON object with:

• "grid": 24x24 list of integers (0–9)
• "palette": list of colors used

Subject: the planet Saturn with its rings

Expected Output Format

{ "grid": [[0,1,1,...],[...],...], "palette": ["black", "yellow", "orange"] }

Constraints:

• grid must be 24×24
• values must be integers (0–9)
• output must be valid JSON

📊 Evaluation Protocol

The benchmark evaluates model outputs across three dimensions, using robust parsing and graded scoring functions.

1. JSON Validity

Evaluates whether the model produces a parsable JSON object, using a multi-stage extraction pipeline:

• Direct JSON parsing
• Regex-based extraction of JSON blocks
• Cleanup of formatting artifacts (e.g., markdown code fences)

Outputs are scored as:

1.0 (correct) → valid JSON extracted 0.0 (incorrect) → parsing failed

2. Render Success (Structural Correctness)

Measures whether the output can be interpreted as a valid 24×24 pixel grid.

This is a graded score ∈ [0,1], composed of:

Height score — how close grid height is to 24 Width score — consistency of row lengths Type score — proportion of valid integer cells

Final score: Render = (height + width + type) / 3

This ensures partial credit for structurally plausible outputs.

3. Pixel Art Quality

Evaluates structural richness and completeness of the generated grid.

This is also a graded score ∈ [0,1], combining:

Color diversity

• number of unique integer values
• scaled with a soft cap at ~8 colors

Grid density

• proportion of valid filled cells in the 24×24 grid

Final score: Quality = 0.7 × diversity + 0.3 × density

Aggregate Score

The final benchmark score is computed as:

Score = 0.4 × JSON Validity + 0.3 × Render Success + 0.3 × Pixel Art Quality