Allow float tile_buffer #405
Conversation
bstadlbauer
force-pushed
the
allow-float-tile-buffer
branch
from
194af4d
to
c36424b
Compare
|
Thanks @bstadlbauer 🙏 I wonder if we could just allow integer and explicitly tell that the value will be Adding a test to check if the tile_buffer is a multiple of 0.5 seems a bit weird TBH. I'll wait for @kylebarron and @geospatial-jeff input |
|
That's also something that I thought about. For me, both versions don't seem quite right so I went with the one not breaking the existing API. I am flexible here, just let me know and I adjust the code as required. |
bstadlbauer
force-pushed
the
allow-float-tile-buffer
branch
from
c36424b
to
8b7ea6d
Compare
vincentsarago
requested review from
kylebarron
and removed request for
kylebarron
|
@bstadlbauer I've rebased this PR to the new rio-tiler-branch and updated the I'll wait for 👍 / 👎 from @geospatial-jeff before merging |
There was a problem hiding this comment.
The code looks great! I wonder if it is more intuitive to change the tile_buffer argument to instead be an absolute buffer. Currently the tile buffer argument is really a padding that is added to all sides of the tile. Instead we could update the argument such that a tile buffer of 1 would result in 257x257 and a tile buffer of 2 would result in 258x258 (assuming your tile size is 256). Obviously this would be a breaking change but maybe that's fine since this will be released with v3.
FWIW I think the current implementation is fine, so please disregard this comment if you don't agree.
|
thanks @geospatial-jeff. We all agree current implementation is fine. I'll add a note in the documentation to make sure users understand what the value of tile_buffer is. |
* switch to morecantile3 (#418) * switch to morecantile3 * remove python 3.6 * make sure to use rio-cogeo2.3.1 in tests * 🤦 * use rio-cogeo from github * update fixtures and tests * update changelog * deprecate metadata methods (#425) * zxy -> xyz in SpatialMixin.tile_exits method (#419) * No max size (#422) * remove default max_size for part and feature * ignore type, failing in python 3.9 * use rio-cogeo alpha * Use RIO_TILER_MAX_THREADS instead of MAX_THREADS (#432) * MAX_THREADS to RIO_TILER_MAX_THREADS * Update env variable in docs * Add to changelog * Update CHANGES.md Co-authored-by: Vincent Sarago <vincent.sarago@gmail.com> * allow non-earth dataset (#429) * allow non-earth dataset * fix stac * metadata/info returns `geographic_bounds` * add test for non earth object tile reading * add notebook * update changelog * update docs * update * revert and remove min/max zoom in __init__ * edit changes * Use httpx (#431) * Replace requests with httpx * Use httpx instead of requests * Use httpx in notebooks * Add tox to dev requirements * Revert change and remove unused import * update changelog Co-authored-by: Vincent Sarago <vincent.sarago@gmail.com> * add new statistics methods and band names in ImageData object (#427) * add new statistics methods and band names in ImageData object * update base classes and tests * MultiBandReader.statistics should use self.preview as COGReader * update tests * start migration docs * update docs * remove `band_expression` option in MultiBandReader (#437) * Asset expression indexes (#438) * change asset_expression type and add asset_indexes * update changelog * moar docs * Allow float tile_buffer (#405) * Add tests of the expected behaviour * Add `tile_buffer` `float` support in `COGReader.tile()` * :facepalm Co-authored-by: Bernhard Stadlbauer <bstadlbauer@blackshark.ai> Co-authored-by: vincentsarago <vincent.sarago@gmail.com> * update docs and allow backward compat for indexes in MultiBaseReader * Range colormap (#439) * add range colormap support * update docs * update types * s/range/intervals/g * remove deprecated code and update docs (#440) * remove deprecated code and update docs * update changelog Co-authored-by: Rodrigo Almeida <rodrigo@developmentseed.org> Co-authored-by: bstadlbauer <11799671+bstadlbauer@users.noreply.github.com> Co-authored-by: Bernhard Stadlbauer <bstadlbauer@blackshark.ai>
Some tile data such a certain digital elevation models (DEMs) default to an uneven tile size (e.g. 257x257). They have an overlap of one pixel on each side, and assume that each pixel is not an "area" but a point measurement. It would be nice, if
COGReader.tile()would support this use case.I found the
tile_bufferparameter, which would do this exact thing. However, as it only supports integers, only overlaps of two pixel are possible. This PR changes the signature oftile_bufferto beOptional[Union[int, float]]where thefloatcase is restricted to halves of integers, to ensure an integertilesize