phelber / eurosat Goto Github PK

Thank you for sharing this great dataset!

When using the three bands of B04, B03 and B02 in MS to make the data of RGB version, I am wondering how the pixel values (DN) of the two versions of data were mapped?
I checked the samples of MS version and RGB version corresponding to the same region. Taking the Red band as an example, the DN value of the MS version sample is not linearly mapped to the RGB version.

Besides, for MS version data, what's your normlization stragery when training with CNNs? Specifically, MS data with pixel values ranging from 0 to 6000 is directly subtracted by A divided by B, where A is a fix number for all 13 bands and whole dataset？

Hi,

I'm looking to overlap the EuroSat dataset with another imagery source. Is it possible to get the id and dates of the Sentinel-2 tiles used for extract the dataset patches. The original Sentinel-2 filenames should suffice.

The ReadMe and paper say it is. The original S2 products are geo-referenced and dated, but the list of the used granules are not provided. but the archive only contains numbered jpeg files, named and classified by the land use class.

Do you have a table indicating the for each filename: the date of capture and coordinates of center or uperright & bottomleft corners ? either in absolute WGS84 or a projection valid on the european zone ? Thank you.

The SSL certificate for https://madm.dfki.de/files/sentinel/EuroSATallBands.zip and https://madm.dfki.de/files/sentinel/EuroSAT.zip has expired, meaning that the dataset can no longer be downloaded. Can someone update this SSL cert?

Thank you for your great paper and dataset!

I would like to know if you have published the GPS coordinate of each observation point? I can't find this metadata in the tiff files. Is it in another file?

Did you use Sentinel-2A Level1-C or Level2-A?
I think the only difference is that the latter is atmospherically corrected.

What is the order of the bands of the MS dataset? Is the 8A band placed as the 13th band? Thank you very much for this great dataset.

---- SOLVED ----
I tried to download the RGB version on different days and with different internet connections. The download page loads but the download does not start.

Hi,

I notice that images in EuroSAT (RGB) are of uint8 datatype with the range [0,256].

However images in EuroSAT (MS) are of uint16 datatype. And the range is different across different bands and different images.

E.g.
AnnualCrop_1.tif band02 [1002,1534], band03 [905,1664]
Forest_1.tif band02 [776,1009]

How do you normalize these multispectral images to the range like uint8 [0,256] or float [0,1]?

Thank you!