MemBench: A Benchmark for Image Memorability Feedback

MemBench is a benchmark dataset for evaluating memorability feedback methods, built on top of the PPR10K photo retouching dataset. It was introduced alongside the MemCoach training-free method by author laitifranz and was last updated on March 3, 2026. The dataset is organized in photoshoots, which are sequences of related images.

Use Cases

• Benchmarking memorability feedback algorithms based on the structured photoshoot sequences.
• Training or evaluating Multimodal LLMs for image memorability tasks using the underlying PPR10K photo retouching data.
• Studying the relationship between image editing (retouching) and perceived memorability.

Strengths

• Built on PPR10K, a large-scale photo retouching dataset.
• Introduced alongside a peer-reviewed method (MemCoach) for memorability feedback.
• Organized in photoshoots, providing contextual sequences of images.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license information are unknown.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

laitifranz on Hugging Face.

Collection Method

Built on top of the PPR10K photo retouching dataset, organized in photoshoot sequences.

Freshness

Last updated 2026-03-03 10:02:54; freshness should be verified.