SNR-Bench: 80 High-Quality Image-Editing Cases

SNR-Bench comprises 80 high-quality image-editing cases. Approximately 50% are sampled from PIE-Bench to ensure continuity with standard benchmarks. The dataset was created by researchers from Zhejiang University's State Key Lab of CAD&CG and submitted for publication in 2026.

Use Cases

• Benchmarking flow-based image editing models based on the 80 provided cases.
• Evaluating structure-aware noise rectification methods on high-quality image pairs.
• Training or fine-tuning generative models for inversion-free editing tasks.
• Comparing performance against established benchmarks like PIE-Bench using the overlapping samples.

Strengths

• Contains 80 high-quality image-editing cases.
• Ensures continuity with standard benchmarks by sampling approximately 50% of its cases from PIE-Bench.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

State Key Lab of CAD&CG, Zhejiang University

Collection Method

Likely curated for research on structure-aware noise rectification for inversion-free flow-based editing.

Freshness

Last updated 2026-04-20 05:33:16; freshness should be verified.