Use ColorInterp or MaskFlags to check alpha band

[kprabowo]

I got this error: https://github.com/mapbox/rasterio/blob/master/rasterio/_warp.pyx#L907

and this code explains that it checks alpha band by GDAL color interpretation.
https://github.com/mapbox/rasterio/blob/master/rasterio/_warp.pyx#L898

p.s: sorry for my bad english

[vincentsarago]

Hi @kn-id, thanks for opening this PR.
To be honest I'm not sure if ColorInterp.alpha == colorinterp: is better than MaskFlags.alpha in flags to check for internal alpha band.

While testing the code I couldn't find a dataset that was passing MaskFlags.alpha in flags but failing ColorInterp.alpha == colorinterp:.

Do you mind sharing your image ? or at least the output of rio info ?

Thanks

[vincentsarago]

also seems @mojodna is using a slightly different test in marblecutter which might be worth to steal copy ;-) https://github.com/mojodna/marblecutter/blob/c7170aa063f5003a6d1acbc95753831d23d5b776/marblecutter/__init__.py#L264-L267

[vincentsarago]

worth noting that

for colorinterp in src.colorinterp:
    if ColorInterp.alpha == colorinterp:
        return True

won't work but should be

if ColorInterp.alpha in src.colorinterp:
    return True

I'm happy to update the function to something like:

def has_alpha_band(src):
    """Check for alpha band or mask in source."""
    if (
        any([MaskFlags.alpha in flags for flags in src.mask_flag_enums])
        or ColorInterp.alpha in src.colorinterp
    ):
        return True
    return False

but I'd love to make sure we understand why the first test (MaskFlags.alpha) didn't work for your image.

[kprabowo]

Hi @vincentsarago, thanks for replying. Here's the output of rio info from my raster image:

{
    "blockxsize": 256,
    "blockysize": 256,
    "bounds": [
        412620.45813000004,
        9667517.052310001,
        414303.85761000006,
        9669577.69789
    ],
    "colorinterp": [
        "red",
        "green",
        "blue",
        "alpha"
    ],
    "compress": "lzw",
    "count": 4,
    "crs": "EPSG:32751",
    "descriptions": [
        null,
        null,
        null,
        null
    ],
    "driver": "GTiff",
    "dtype": "uint8",
    "height": 14721,
    "indexes": [
        1,
        2,
        3,
        4
    ],
    "interleave": "pixel",
    "lnglat": [
        122.22127123720861,
        -2.998450073465773
    ],
    "mask_flags": [
        [
            "nodata"
        ],
        [
            "nodata"
        ],
        [
            "nodata"
        ],
        [
            "nodata"
        ]
    ],
    "nodata": null,
    "res": [
        0.13998000000000002,
        0.13998000000000002
    ],
    "shape": [
        14721,
        12026
    ],
    "tiled": true,
    "transform": [
        0.13998000000000002,
        0.0,
        412620.45813000004,
        0.0,
        -0.13998000000000002,
        9669577.69789,
        0.0,
        0.0,
        1.0
    ],
    "units": [
        null,
        null,
        null,
        null
    ],
    "width": 12026
}

And here's the link to my image https://drive.google.com/file/d/1pJdJ6sdNUQLrOPn4AZqaHoSN_zJrFihS/view?usp=sharing

[vincentsarago]

having mask_flags=Nodata and colorinterp=alpha is kind of bizarre. Any idea how the file was created ?

[kprabowo]

It was created using pix4d. Any idea how to solve this?
should i update the code according to your suggestion?

def has_alpha_band(src):
    """Check for alpha band or mask in source."""
    if (
        any([MaskFlags.alpha in flags for flags in src.mask_flag_enums])
        or ColorInterp.alpha in src.colorinterp
    ):
        return True
    return False

[vincentsarago]

pix4d its really doing something weird, we already talked about it here
rasterio/rasterio#1373 (comment)

I have no objection to update the code with the proposed change
#51 (comment) but just noting that having the nodata value ignore in the file is kind of unwanted, this should be reported to pix4d.

[kprabowo]

Hi @vincentsarago, the code has been updated, please review this. Thank you.

[vincentsarago]

👍 thanks @kn-id

[vincentsarago]

@kn-id I'll add a test case for this new code and publish a new version of rio-tiler later today

[vincentsarago]

Hi @kn-id, I cannot create a file to test the new code (to add the test fixture). I tried to use gdal_translate In.tif out.tif -outsize 3% 3% to get a < 1Mb file, but the file doesn't behave like the initial file from pix4d.
Would you be able to create a small file directly from pix4d ?

[kprabowo]

@vincentsarago , I'm only able to get as small as 1,5 Mb. Here's the link https://drive.google.com/open?id=0Bzo9qLZh8_--c0RsS2l0WnBzbW1ya2RWZFlZNjFCZzZRUTRR
Is this okay?

[vincentsarago]

thanks @kn-id, just release rio-tiler==1.0rc2 with the fix 🙏