license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) solve mask version 8
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data.jsonl
Version 1
ARC-AGI Tasks where the job is to repair the masked areas/rectangles.
example count: 2-4.
test count: 1-2.
image size: 4-7.
noise: 0.1, 0.2.
There are these transformations: identify_the_masked_area, repair_the_masked_area
Version 2
image size: 4-10.
Version 3
image size: 4-13.
Version 4
Still having all the other transformations enabled.
Added generate_task_repair_rectangle_and_crop.
input image size: 4-8.
mask size: 2-3.
Version 5
Bigger images.
generate_task_repair_rectangle_and_crop: image size: 4-10. crop size: 2-4.
generate_task_linepatterns_with_masked_areas: image size: 4-15.
Version 6
Earlier predictions added to some of the rows.
Smaller images. generate_task_repair_rectangle_and_crop: image size: 4-8. crop size: 2-4.
Smaller images. generate_task_linepatterns_with_masked_areas: image size: 4-12.
Version 7
Added fields: arc_task, test_index, earlier_output.
Version 8
Replaced RLE compressed response with raw pixel response.
Big images. generate_task_repair_rectangle_and_crop: image size: 4-10. crop size: 2-6.
Big images. generate_task_linepatterns_with_masked_areas: image size: 4-15.