ComplexCompo: A Benchmark for Evaluating Image Composition Methods

ComplexCompo is a benchmark dataset constructed by Lian Zhuming for evaluating image composition methods in terms of physical plausibility. It integrates foreground objects from DreamEditBench and background images from Open Images V7, and introduces newly annotated bounding boxes, prompts, and a composition pairing protocol. The dataset was last updated on February 15, 2026.

Use Cases

• Benchmarking physical plausibility of image composition models based on the described pairing protocol.
• Training models for realistic object insertion using the annotated foreground and background pairs.
• Evaluating the quality of generated prompts for image editing tasks based on the provided annotations.

Strengths

• Constructed specifically for evaluating physical plausibility in image composition, a defined research goal.
• Integrates data from two established sources: DreamEditBench for foregrounds and Open Images V7 for backgrounds.
• Includes newly annotated bounding boxes and prompts, adding value beyond the source datasets.

Limitations

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

Provenance

Source

Foreground objects from DreamEditBench, background images from Open Images V7.

Collection Method

Constructed by integrating and annotating existing datasets with a new composition pairing protocol.

Freshness

Last updated 2026-02-15 08:25:25; freshness should be verified.