SOLA (Simulated Objects for Language Acquisition) is a small clean dataset with little noise and clearly defined attributes for efficient comparative learning and grounded language acquisition. It is generated using the open-source simulation software Kubric.
Dataset Stats
| Learning Attributes | Changing Attributes | Variation Attributes |
|---|---|---|
| Color: 8 | Lighting: 3 | Shade: 3 |
| Material: 4 | Camera Angle: 6 | Size: 3 |
| Shape: 11 | Stretch: 4 | |
| Total: | 6336 (RBGA) | 989 (RGBA) |
Image Types (5): RGBA, Depth, Surface Normal, Segmentation Map, Object Coordinates