---
license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) shape version 5
size_categories:
- 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data.jsonl
---

# Version 1

Detect `shape2x2` and `shape3x3_center`.
The image sizes are between 1 and 30 pixels.

# Version 2

Detect `shape2x2` and `shape3x3_center` and `shape3x3_opposite`.
The image sizes are between 1 and 30 pixels.

# Version 3

Focus on counting the unique number of colors. `corners` and `diamond4`.
The image sizes are between 1 and 30 pixels.

# Version 4

Same weight to all transformations.
The image sizes are between 1 and 30 pixels.

# Version 5

Added more ways of counting the unique number of colors. `all9`, `around_center` and `diamond5`.
Same weight to all transformations.
The image sizes are between 1 and 30 pixels.