Prox-E ShapeTalk Benchmark is a subset of 600 random samples from the ShapeTalk dataset, curated for evaluating 3D shape editing models. It is the official benchmark for the SIGGRAPH'26 paper 'Prox-E: Fine-Grained 3D Shape Editing via Primitive-Based Abstractions'. The dataset was created by author 'haopt' and last updated on May 30, 2026.
Use Cases
- Benchmarking identity preservation in 3D shape editing based on the described evaluation purpose
- Evaluating 3D quality of generated shapes based on the dataset's stated focus
- Assessing edit fidelity of shape manipulation models based on the benchmark's intended use
- Training or fine-tuning 3D generative models on a curated subset of shape-language data
Strengths
- Contains 600 samples, providing a defined scale for benchmarking
- Serves as the official benchmark for a SIGGRAPH'26 publication, indicating academic relevance
Limitations
- Column-level documentation is absent; field semantics must be inferred after download
- Row count is unknown beyond the 600-sample subset, which may limit suitability assessment
Provenance
- Source
- Subset of the ShapeTalk dataset
- Collection Method
- Random sampling of 600 samples
- Freshness
- Last updated 2026-05-30 19:57:52