You can check how a S2 COG compares at https://registry.opendata.aws/sentinel-2-l2a-cogs/
You should hopefully see a lot less network requests when accessing a sub block directly. P On Mon, Nov 2, 2020 at 7:58 AM Alexander Verbitsky <habibu...@gmail.com> wrote: > Hi All, I am trying to implement remote reading small blocks of images, > but find slow performance. > > For example i am trying to read 100x100 block from `sentinel-2` image, > with using following code: > > filename = ( > '/vsigs/gcp-public-data-sentinel-2/tiles/31/U/FS/' > 'S2A_MSIL1C_20201024T104121_N0209_R008_T31UFS_20201024T125115.SAFE/' > > 'GRANULE/L1C_T31UFS_A027890_20201024T104121/IMG_DATA/T31UFS_20201024T104121_B08.jp2' > ) > ds = gdal.OpenEx(filename, gdal.OF_RASTER) > band = ds.GetRasterBand(1) > img = band.ReadAsArray(1000, 1000, 100, 100) > > In result I am getting more low performance in comparison if I just > download the whole image in memory and then read the necessary block. AS I > understand it happens due to a lot of small requests (is about 100) that > are made during reading a small part of the image. Is it a fundamental > feature of JP2OpenJPEG and this format not suitable for doing things like > that? Should I take another format like GeoTiff and store my own data in > it? Thank, for your reply. > > > _______________________________________________ > gdal-dev mailing list > gdal-dev@lists.osgeo.org > https://lists.osgeo.org/mailman/listinfo/gdal-dev
_______________________________________________ gdal-dev mailing list gdal-dev@lists.osgeo.org https://lists.osgeo.org/mailman/listinfo/gdal-dev
