Statistics of the 3D Generation Skills
This section presents detailed statistics for skills categorized under the 3D modality.
The 3D generation tasks emphasize shape synthesis, reconstruction, and completion from sparse or degraded geometric inputs.
These tasks require spatial reasoning and generative modeling capabilities and are evaluated using geometry-aware metrics such as Chamfer Distance and Earth Mover’s Distance.
D-G-1 Point Cloud Completion (PC Complt)
This cluster focuses on completing incomplete 3D point clouds by predicting missing geometry.
- 3D Point Cloud Completion
This task completes partial 3D point clouds into full object shapes.
- Abbreviation:
3D-PC Completion
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
PCN data
- Number:
500
- SoTA Specialist:
PointTr
- Metrics:
CD
D-G-2 Point Cloud to Mesh Reconstruction (PC2M Recon)
This cluster focuses on reconstructing full mesh objects or scenes from point cloud data.
- Point Cloud to Mesh Object Reconstruction
This task reconstructs mesh models of individual objects from point cloud input.
- Abbreviation:
Object-PC2M Recon
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
ShapeNet
- Number:
500
- SoTA Specialist:
ConvONet
- Metrics:
CD
- Point Cloud to Mesh Scene Reconstruction
This task reconstructs full mesh scenes from indoor point cloud scans.
- Abbreviation:
Scene-PC2M Recon
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Synthetic Indoor Scene Dataset
- Number:
500
- SoTA Specialist:
ConvONet
- Metrics:
CD
D-G-3 Text to Point Cloud Generation (Txt2PC Gen)
This cluster focuses on generating 3D point clouds from textual descriptions.
- Text to 3D Living and Arts Point Cloud Generation
This task generates artistic or daily-life 3D shapes from text.
- Abbreviation:
Art-Txt2PC Gen
- Domain:
Art
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Text to 3D Science and Technology Point Cloud Generation
This task generates point clouds related to science and tech concepts.
- Abbreviation:
Sci-Txt2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Text to 3D Nature and Biology Point Cloud Generation
This task creates 3D shapes based on nature- and biology-related prompts.
- Abbreviation:
Nature-Txt2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Text to 3D Culture and Structure Point Cloud Generation
This task focuses on generating culturally or architecturally meaningful 3D forms.
- Abbreviation:
Culture-Txt2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
D-G-4 Text to Mesh Generation (Txt2M Gen)
This cluster focuses on generating 3D mesh representations from text prompts in various domains.
- Text to 3D Living and Arts Mesh Generation
This task generates artistic or everyday 3D mesh models from textual input.
- Abbreviation:
Arts-Txt2M Gen
- Domain:
Art
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Text to 3D Science and Technology Mesh Generation
This task generates mesh shapes related to science and technology topics.
- Abbreviation:
Sci-Txt2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Text to 3D Nature and Biology Mesh Generation
This task generates mesh models of natural or biological entities from text.
- Abbreviation:
Nature-Txt2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Text to 3D Culture and Structure Mesh Generation
This task focuses on mesh generation of cultural artifacts or structures via text.
- Abbreviation:
Culture-Txt2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Manual
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
D-G-5 Image to Point Cloud Generation (Img2PC Gen)
This cluster focuses on generating 3D point clouds from 2D images across various thematic domains.
- Living and Arts Image to 3D Point Cloud Generation
This task generates 3D point clouds of artistic or everyday objects from images.
- Abbreviation:
Arts-Img2PC Gen
- Domain:
Art
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Science and Technology Image to 3D Point Cloud Generation
This task generates point clouds from images related to scientific or technical items.
- Abbreviation:
Sci-Img2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Nature and Biology Image to 3D Point Cloud Generation
This task creates biological or natural object point clouds from images.
- Abbreviation:
Nature-Img2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
- Culture and Structure Image to 3D Point Cloud Generation
This task generates structural or culturally meaningful 3D shapes from image inputs.
- Abbreviation:
Culture-Img2PC Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Point-E
- Metrics:
CLIP-Score
D-G-6 Image to Mesh Generation (Img2M Gen)
This cluster focuses on generating 3D mesh models from 2D image inputs across various domains.
- Living and Arts Image to 3D Mesh Generation
This task generates artistic or daily-use 3D mesh models from images.
- Abbreviation:
Arts-Img2M Gen
- Domain:
Art
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Science and Technology Image to 3D Mesh Generation
This task produces scientific and technical object meshes from image inputs.
- Abbreviation:
Sci-Img2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Nature and Biology Image to 3D Mesh Generation
This task generates mesh representations of biological or natural objects.
- Abbreviation:
Nature-Img2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
- Culture and Structure Image to 3D Mesh Generation
This task converts images of cultural or architectural subjects into mesh models.
- Abbreviation:
Culture-Img2M Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
Objaverse
- Number:
500
- SoTA Specialist:
Hunyuan3D-1.0
- Metrics:
CLIP-Score
D-G-7 RGB-D to Point Cloud Reconstruction (RGBD2PC Recon)
This cluster focuses on reconstructing full 3D point clouds from RGB-D (color + depth) image inputs.
- RGB-D to Point Cloud Reconstruction
This task reconstructs 3D point clouds from RGB-D image data.
- Abbreviation:
RGBD2PC Recon
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
ScanNet
- Number:
142
- SoTA Specialist:
open3d
- Metrics:
CD
D-G-8 RGB-D to Mesh Reconstruction (RGBD2Mesh Recon)
This cluster focuses on producing complete 3D meshes from RGB-D input signals.
- RGB-D to Mesh Reconstruction
This task reconstructs watertight mesh surfaces from RGB-D frames.
- Abbreviation:
RGBD2Mesh Recon
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
ScanNet
- Number:
142
- SoTA Specialist:
open3d
- Metrics:
CD
D-G-9 Text to 3D Motion Generation (Txt2Motion Gen)
This cluster focuses on synthesizing human or object motion in 3D space from text prompts.
- Text to 3D Motion Generation
This task generates plausible 3D motion sequences conditioned on natural language instructions.
- Abbreviation:
Txt2Motion Gen
- Domain:
General
- Capability:
Creativity and Innovation
- Data Source:
KIT-ML
- Number:
830
- SoTA Specialist:
MoMask
- Metrics:
FID