Hillshade algorithm showing grids instead

[lawrencenika]

Problem description

Tried with /cog/tiles/WebMercatorQuad/{z}/{x}/{y}.png?url=${encodeURIComponent(objPath)}&rescale=127.0,236.0&algorithm=hillshade&buffer=3 endpoint and parameters with my current COG data to render on DeckGL map's BitmapLayer. I got the grid looking image tiles instead of terrain looking ones:

High zoom:

Low zoom:

Expected Output

Expected look on other platforms's hillshade algorithm:

Environment Information

titiler:latest (v0.18.6) run in local docker env on macos
with ENV WORKERS_PER_CORE 1

[vincentsarago]

do not rescale the data

when passing the rescale=127.0,236.0 you'll tell titiler rescale the output image from the algorithm, which in the hillshade case, should already be a valid uint8 dataset

https://github.com/developmentseed/titiler/blob/main/src/titiler/core/titiler/core/factory.py#L615-L619

[lawrencenika]

still same grid when not using rescale at all, could it be because my COG raw image is of float32? https://drive.google.com/file/d/1bZQAvxEBbPMyjkXhNYEVyQQck0dNMFRx/view?usp=sharing

[lawrencenika]

@vincentsarago this one too

[vincentsarago]

@lawrencenika you're file is not a DEM right? it's a 4 bands dataset of type float

The hillshade algorithm is defined here

titiler/src/titiler/core/titiler/core/algorithm/dem.py

Lines 31 to 39 in 725f857

	x, y = numpy.gradient(img.array[0])
	slope = numpy.pi / 2.0 - numpy.arctan(numpy.sqrt(x * x + y * y))
	aspect = numpy.arctan2(-x, y)
	azimuthrad = self.azimuth * numpy.pi / 180.0
	altituderad = self.angle_altitude * numpy.pi / 180.0
	shaded = numpy.sin(altituderad) * numpy.sin(slope) + numpy.cos(
	altituderad
	) * numpy.cos(slope) * numpy.cos(azimuthrad - aspect)
	data = 255 * (shaded + 1) / 2

The code is pretty basic and used to work for DEM, I'm not sure what Is wrong with the usage you're trying to do.

you're sharing an expected output, can I ask how you created it?

[lawrencenika]

yes my image is not DEM but 4 bands of type float. the created output are shown as images above expected output.

[vincentsarago]

Ok I think there is something going on with rio-tiler when we are oversampling a file somewhere https://github.com/cogeotiff/rio-tiler/blob/741acc3596d908db1755c82f18e2f4fd3711c49c/rio_tiler/reader.py#L375-L423

I'm not quite sure why and where but I'll open an issue in rio-tiler

[lawrencenika]

thanks, awesome!

[vincentsarago]

@lawrencenika I'm still looking at this.

I already found one issue which will be fixed in #985 and I'll backport this in a non-breaking version (0.18.7, released later today)

The other main issue comes from rio-tiler, when you're over-zooming your dataset, which for some reason creates the square artefacts. For now I don't see immediate fix for this, so it would be best if you don't fetch tiles that are over-zooming you're dataset and stop at the maxzoom returned in the tilejson endpoint.

Your map client should be able to over-zoom the pre-loaded tiles them self

[lawrencenika]

awesome to hear! would the rio-tiler issue solved later avoid having to use the tilejson endpoint?

[vincentsarago]

I don't really have the bandwidth at the moment. And it's a good habit not to over zoom a dataset anyway!

[lawrencenika]

got it thanks!