Replies: 3 comments 1 reply
-
My personal opinion is that all COGs should be tiled even if they are small. The storage overhead of writing extra tiff headers is so small it doesn't matter too much (storage is cheap), especially considering most readers aim to fetch the entire TIFF header in one range request (usually ~16-64kb). COG compliant software (readers / writers / validators etc.) will be much simpler if they only have to support tiled images, instead of tiled and strip. |
Beta Was this translation helpful? Give feedback.
-
rasterio has just been updated: rasterio/rasterio#2615 and it changes everything. I'll update rio-cogeo |
Beta Was this translation helpful? Give feedback.
-
This should now be fixed. with rio-cogeo>=3.5 |
Beta Was this translation helpful? Give feedback.
-
👋 hi all,
I am curious to dig into the
IncompatibleBlockRasterSize
check that exists inrio-cogeo
for small 2D dimension files (e.g. 34x29 px)I have run across tools (in my case Google EarthEngine) that are only happy to accept COGs, but check if the file is Tiled as part of COG validation. This will fail if
rio cogeo create
with default settings as it will not write blocks due to the dimension size check, and thus the file reports inrio cogeo info
as a valid COG, butFalse
under Tiled. If you specify a block size, you getIncompatibleBlockRasterSize
exceptions.GDAL does not seem to care, and if you
gdal_translate --of COG
it will write a TIFF w/ the default BlockSize, extra space on disk required and all.My understanding has always been that to lay your data out in a tiled format, you have to write more TiffTags which may end up costing more in storage sapce, when writing data with small dimensions (e.g. 32 x 32 px). Should a COG be considered valid if it is not Tiled? Is there a reason why to enforce skipping writing in a tiled format even for small images?
Curious to hear what others thoughts are
Beta Was this translation helpful? Give feedback.
All reactions