license: mit
task_categories:
- image-to-text
- text-to-image
language:
- en
pretty_name: simons ARC (abstraction & reasoning corpus) solve boundingbox version 16
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 identify the boundingbox.
example count: 2-4.
test count: 1-2.
image size: 3-8.
filled bounding box.
Version 2
image size: 3-15.
filled bounding box.
Version 3
image size: 3-15.
filled+hollow bounding box.
Version 4
image size: 3-20.
filled+hollow bounding box.
Version 5
image size: 3-30.
filled+hollow bounding box.
Version 6
image size: 3-30.
Added more variants: filled_inner, filled_outer, hollow_inner, hollow_outer.
Version 7
Earlier predictions added to some of the rows.
Version 8
Added fields: arc_task, test_index, earlier_output.
Version 9
Added generate_task_inner_boundingbox where it's about find the inner bounding box.
max image size = 12.
inner pattern size = 2-3.
Version 10
Bigger images with generate_task_inner_boundingbox.
max image size = 14.
inner pattern size = 2-4.
Version 11
Bigger images with generate_task_inner_boundingbox.
max image size = 16.
inner pattern size = 2-5.
Version 12
Bigger images with generate_task_inner_boundingbox.
max image size = 18.
inner pattern size = 2-5.
Version 13
Extended generate_task_inner_boundingbox with swap_background_outer transformation.
max image size = 10.
inner pattern size = 2-3.
Version 14
Focus on swap_background_outer transformation.
max image size = 12.
inner pattern size = 2-4.
Version 15
Focus on swap_background_outer transformation.
max image size = 14.
inner pattern size = 2-6.
Version 16
Focus on swap_background_outer transformation.
max image size = 16.
inner pattern size = 2-6.