Tropical Andes Land Cover Dataset (TALANDCOVER)

The Tropical Andes Land Cover Dataset (TALANDCOVER) was built for the department of Antioquia Colombia and consists of three folders for the three types of sampling.

Random: 5000 images
Balanced of minimum 50% coverage per class: 1389 images
Balanced of minimum 70% coverage per class:731 images
The coordinate system of each dataset is EPSG:3857 - WGS 84 / Pseudo-Mercator with spatial resolution of 4.77 meters and 128*128px.
Example of image and corresponding label name:

image_PNICFI_D2019-05_T586-1068_C1_N100.tif
label_PNICFI_D2019-05_T586-1068_C1_N100.tif
Pixel values for label are:
0 Bare-degraded lands1 Grasslands2 Heterogeneous agricultural areas3 Dense forest4 Water bodies5 Built-up areas
There are multiple keywords intentionally inserted in each image name or label name that enable the split of the name in pieces of information about each image and label metadata.
Spliting the image or label name by the keywords, would get 6 items:

The item (image/label) describe if the file correspond to a patch image or a patch label.
The second item (Keyword "_P") gives the name of the product NICFI (https://assets.planet.com/docs/NICFI_User_Guide_v4_EN.pdf)
The third item (keyword "_D") gives a date for the composite
The fourth (keyword "_T") gives the tile number, as stated in the planet scope visual base map documentation: "The name of each basemap quad within the Basemaps API is designed to represent the x and y position of the quad within the two dimensional grid which makes up the basemap. It is generally {X}-{Y}, where X and Y are the x and y position of the quad in the grid". https://developers.planet.com/docs/data/visual-basemaps/
The fifth (keyword "_C") when available gives the cover class used by the slidding windows to extract the patch, resulting in at least a minimum of 50% or 70% of the pixels within the image correspond to that specific cover class, depending on the selected sample dataset.
Take into account that random samples dont have this keyword since the patches where collected at random from a 2d grid without regard of the cover classes present

Article : "Land Cover Classification in the Antioquia Region of the Tropical Andes Using NICFI Satellite Data Program Imagery and Semantic Segmentation Techniques".