metadata

configs:
  - config_name: v2
    data_files:
      - split: train
        path: v2/train_color.h5
      - split: test
        path: v2/val_color.h5
  - config_name: v1
    data_files:
      - split: train
        path: v1/train_color.h5
      - split: test
        path: v1/val_color.h5
  - config_name: mini_test
    data_files:
      - split: train
        path: mini_test/train_color.h5
      - split: test
        path: mini_test/val_color.h5
license: mit
task_categories:
  - image-to-image

scnn-watermark-removal dataset for classified house cleaning ads

name	tn	ts	val	size	scales*	augm**	release
test	1	77	1	69	4	yes	2024-12-01
v1	477	9048	50	7,575	?	no	2024-12-01
v2	48	2179	4	1,732	4	no	2024-12-01
v3	48	876	6	723	1	no	2024-12-01

* 1: [1]; 4: [1, 0.9, 0.8, 0.7] ** # of augmentations performed on each patch

Dataset Composition

Source Portals:

ME Medellín
MR Colombia
KM Germany
OL International

Exclusion Criteria:

Watermark present
One dimension below 250 px

Post Processing:

For most datasets width has been downsized to 460 px while maintaining aspect ratio

Total Collection (2024-12-01)

Source	# Samples
ME Medellín	276
MR Colombia	1,154
KM Germany	46
OL International	153

mini_test

* train picture not suited for real training. The purpose of this version is to test if the actual training will work