license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) solve translate version 11
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data.jsonl
Version 1
ARC-AGI Tasks where the image gets translated by plus/minus 1 pixel in up/down/left/right directions.
The image sizes are between 1 and 4 pixels.
Version 2
Only translate plus/minus 1 up/down are enabled.
image width: 1-4, image height: 3-4.
My hypothesis is that it's easy with RLE data to translate up/down.
Version 3
Only translate plus/minus 1 left/right are enabled.
image width: 3-4, image height: 1-4.
Version 4
All transformations have same weight.
image size: 3-4.
Version 5
Added diagonal translation by 1 pixel.
All transformations have same weight.
image size: 3-4.
Version 6
All transformations have same weight.
image size: 3-5.
Version 7
All transformations have same weight.
image size: 3-5.
number of test pairs: 1-2. Previous it was alway 1 test pair.
Version 8
All transformations have same weight.
image size: 3-5.
number of test pairs: 1-2.
Added: Predict the number of rows in the output image.
Version 9
Increased the translation distance from -1..+1, to -2..+2.
image size 1-8.
Version 10
Increased the translation distance from -2..+2, to -3..+3.
image size 1-12.
Version 11
Added fields: arc_task, test_index, earlier_output.