RevealLayer 100K: Box-Guided Layered Image Decomposition for Natural Scenes

RevealLayer Open is the open-source dataset accompanying the ICML 2026 paper 'RevealLayer: Disentangling Hidden and Visible Layers via Occlusion-Aware Image Decomposition'. The dataset supports research into box-guided layered image decomposition for natural images. It was created by qihoo360 and last updated on Hugging Face in May 2026.

Use Cases

• Training models for layered scene decomposition based on instance bounding boxes mentioned in the description.
• Research into occlusion-aware image editing and composition based on the described foreground/background separation task.
• Benchmarking instance-level image decomposition algorithms based on the described box-guided approach.

Strengths

• Dataset is associated with a peer-reviewed paper accepted by ICML 2026.
• The description specifies a clear task: decomposing scenes into a clean background and object-level foreground layers.

Limitations

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

Provenance

Source

qihoo360

Collection Method

Created to accompany the research paper 'RevealLayer: Disentangling Hidden and Visible Layers via Occlusion-Aware Image Decomposition'.

Freshness

Last updated 2026-05-18 12:59:10; freshness should be verified.