Edge Quality Benchmark: Hair & Fur — SimplyPNG v1.1.0

Benchmark Status

Quantitative scores will be added to this page after the first benchmark run completes. The methodology is finalized and documented below.

Why Hair & Fur Are the Hardest Cases

Hair and fur represent the maximum difficulty in background removal because they combine three challenging properties simultaneously:

1 — Sub-pixel strand width

Individual hair strands in a typical portrait photograph (24MP camera, cropped to 3000px) are approximately 1–3 pixels wide. At this scale, most pixels along the strand boundary are "mixed" — containing color information from both the strand and the background. Correctly predicting the fractional contribution of each color requires alpha matting rather than binary segmentation.

2 — Strand overlap and layering

Hair does not have clean, parallel edges. Strands cross each other, create depth, and form irregular profiles at the hairline. The AI must simultaneously handle hundreds of independent boundaries, each with different orientations and backgrounds.

3 — Fine flyaway strands

Loose strands that extend away from the main hairline are often only 1 pixel wide and may be separated from the main hair mass by background pixels. These require the model to detect and preserve isolated thin features — a harder problem than continuous edge refinement.

Methodology

Quality Scale

Inter-rater reliability is measured with Cohen's kappa (κ). Target: κ ≥ 0.75 (substantial agreement).

Test Image Set

Each image processed in Fast mode (2500px, 1 credit) and HD mode (4096px, 2 credits) — 80 total runs.

Processing Setup

• API version: SimplyPNG v1 /api/v1/jobs
• Output: PNG with transparent background (preserves alpha channel for accurate edge inspection)
• Rating method: Each rater evaluates the output independently; scores are averaged after rating is complete
• Processing: Test images submitted via SimplyPNG API with transparent output; outputs downloaded for human rating

Fast Mode vs HD Mode: What to Expect

The primary difference for hair and fur edges is the number of pixels available per strand:

Recommendation: Use HD mode for any image where individual hair or fur strands are visible at the final display size and matter for the end use case.

Tips for Best Hair & Fur Results

1. Shoot at maximum resolution — The AI can only preserve detail that exists in the source. Shoot at the highest practical camera resolution and avoid aggressive in-camera JPEG sharpening.

2. Use contrasting backgrounds when shooting — High contrast between hair and background (dark hair on white, light hair on dark) gives the AI more signal to work with. Avoid shooting blonde hair against cream or white backgrounds.

3. Choose HD mode for commercial hero shots — For images that will appear at full size on product pages or in advertising, HD mode's full-resolution processing preserves the fine strand detail that makes the output look natural rather than cut out.

4. Export as PNG when the output will be composited — PNG preserves the alpha channel (fractional opacity per pixel) that makes hair edges look soft and natural when placed on a new background. JPEG forces a hard edge.

5. Avoid heavy background blur (bokeh) — While bokeh backgrounds look good in photography, heavy blur at the hair boundary reduces contrast and makes strand-level separation harder. A slightly out-of-focus but recognizable background is easier to remove than a solid-tone blur that blends with hair color.

Related Resources

• White Background Accuracy Benchmark — BM-WB-001
• SimplyPNG vs remove.bg — Feature comparison
• SimplyPNG vs Canva — Feature comparison
• E-Commerce Image Compliance Guide — Marketplace requirements

Benchmark ID: BM-EQ-001 | Version: SimplyPNG v1.1.0 | Updated: February 25, 2026 Quantitative results: Scheduled Q1 2026 — this page will be updated when the benchmark run completes.