Edit3D-Bench: 300 Text-Based Editing Prompts for 100 3D Models

Edit3D-Bench consists of 100 high-quality 3D models and 300 editing prompts designed for evaluating 3D editing algorithms. Developed by huanngzh for the 2025 VoxHammer paper, the collection sources 50 models from Google Scanned Objects and 50 from PartObjaverse-Tiny.

Use Cases

• Benchmarking text-to-3D editing performance using the 3 distinct editing prompts per model
• Evaluating part-specific modifications on the PartObjaverse-Tiny subset
• Validating 3D asset generation against the provided annotated ground truth assets

Strengths

• 300 unique editing scenarios
• Includes high-quality source data from Google Scanned Objects
• Provides annotated 3D assets for ground truth comparison

Limitations

• Small sample size of 100 base models
• Potential domain bias toward objects found in GSO and PartObjaverse-Tiny
• Unknown file formats for the 3D assets

Provenance

Source

VoxHammer: Training-Free Precise and Coherent 3D Editing in Native 3D Space (Arxiv 2508.19247)

Collection Method

Curated from Google Scanned Objects and PartObjaverse-Tiny with manual prompt annotation

Freshness

Last updated September 2025.