AutoRIFT assumes single-band optical data by default (?)

[ehultee]

Hi there!

I am playing with some drone data and hoping to run it through autoRIFT for velocities. It is optical data with 4 bands, saved by rasterio with shape (4, 30921, 36965). I am getting an error that suggests AutoRIFT is not expecting this structure:

>> python testautoRIFT.py -m tif1_CommonSample.tif -s tif2_CommonSample.tif -fo 1

/Users/eultee/miniforge3/envs/arift/lib/python3.12/site-packages/osgeo/gdal.py:311: FutureWarning: Neither gdal.UseExceptions() nor gdal.DontUseExceptions() has been explicitly called. In GDAL 4.0, exceptions will be enabled by default.
  warnings.warn(
Preprocessing filter width 5
Using preprocessing methods ['hps', 'hps']
Setting Wallis Filter Width to 5
Traceback (most recent call last):
  File "/Users/eultee/Documents/GitHub/autoRIFT/testautoRIFT.py", line 1137, in <module>
    main()
  File "/Users/eultee/Documents/GitHub/autoRIFT/testautoRIFT.py", line 411, in main
    generateAutoriftProduct(indir_m=inps.indir_m, indir_s=inps.indir_s, grid_location=inps.grid_location,
  File "/Users/eultee/Documents/GitHub/autoRIFT/testautoRIFT.py", line 545, in generateAutoriftProduct
    runAutorift(
  File "/Users/eultee/Documents/GitHub/autoRIFT/testautoRIFT.py", line 188, in runAutorift
    m,n = obj.I1.shape
    ^^^
ValueError: too many values to unpack (expected 2)

When I read I1 using GDAL's ReadAsArray, its shape is (4, 30921, 36965) -- as I would expect. But it appears that loadProductOptical expects the output of read as array to have shape (m,n) rather than (nbands, m, n).

One fix I can think of is to apply loadProductOptical to read in each band of my data (e.g. using GDAL’s band=DS.GetRasterBand(1)) instead of the whole image. But then I suppose I have to loop over all the bands, and somehow reassemble them? I see that other functions in autoRIFT are aware of bands.

I am not very experienced in this level of RS processing, so if there’s a usual community practice I’m missing here, I’m happy to hear about it (and sorry to raise an issue in that case!). And if I happen to find a workaround with loadProductOptical, I'll report back. Happy for other thoughts in the meantime!

[jhkennedy]

Hi @ehultee! For Sentinel-2 and Landsat, we're usually processing a specific band, and each band comes as a separate tiff. What data/mission are you trying to process?

For a multi-band image, what space are you expecting to do the cross-correlation(s) in? That is, would you like offsets calculated for each band, or would you like to apply the cross-correlation to a decomposition of the bands, or do the cross-correlation across all bands simultaneously (nbands-dimentional correlations)?