Sign in to view source links and access this dataset
Description
Pxform_v1 is a part-level, instruction-based 3D editing dataset. Each record contains a (before, after) pair of 3D SLAT latents with aligned RGB views, annotated by a natural-language edit prompt. The dataset covers seven edit types, including deletion, addition, modification, scale, material, and color.
Use Cases
Train models for 3D object deletion based on natural-language prompts.
Develop systems for 3D object addition guided by textual instructions.
Fine-tune models for modifying 3D object properties like scale, material, or color via language.
Benchmark the performance of multimodal AI systems on part-level 3D editing tasks.
Strengths
Dataset is structured around seven specific edit types, providing clear task definitions.
Each data point includes aligned RGB views for both before and after states, offering visual context.
Data pairs are annotated with natural-language edit prompts, enabling instruction-based learning.
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
ART-3D
Freshness
Last updated 2026-05-26 17:46:51; freshness should be verified.
License is unknown; users should verify terms before use.