Synthetic Image Dataset for Geometry-Aware Spatial Editing

SpatialEdit-500K is a synthetic training dataset containing 500,000 entries for fine-grained image spatial editing. It was created by EasonXiao-888 using a controllable rendering pipeline to generate structured spatial transformations and was last updated in April 2026.

Use Cases

• Training models to perform object moving based on provided spatial transformation parameters.
• Learning object rotation edits using the dataset's structured geometric annotations.
• Simulating camera viewpoint changes by leveraging the controllable rendering pipeline outputs.
• Developing models for fine-grained, spatially grounded image editing tasks.

Strengths

• Contains 500,000 synthetic image-edit pairs for large-scale training.
• Provides structured spatial transformations for geometry-aware learning.

Limitations

• Dataset is entirely synthetic, which may limit real-world applicability and generalization.
• Specific row counts, column details, and file formats are not provided in the description.

Provenance

Source

EasonXiao-888 on Hugging Face.

Collection Method

Generated with a controllable rendering pipeline as part of the SpatialEdit project.

Freshness

Last updated on 2026-04-06.