Implement Nan padding in blockwise coregistration stats

[vschaffn]

Resolves #604

Description

This PR introduces several bug fixes to the BlockwiseCoreg functionnality, focusing on handling missing data and improving robustness in the stats method.

Key changes

In the section of the code where the statistics are collected, for chunks that fail (i.e., where points are not present in chunk._meta), the center coordinates are retrieved, and NaN values are now stored for the relevant statistics, replacing the absence of values.

Example of the BlockwiseCoreg.stats() method output with failing chunks 0 and 1:

chunk	center_x	center_y	center_z	x_off	y_off	z_off	inlier_count	nmad	median
0	1.131960e+06	-644298.185858	NaN	NaN	NaN	NaN	NaN	NaN	NaN
1	1.132980e+06	-644298.185858	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2	1.132080e+06	-644358.185858	152.551422	2.563047	8.209118	-4.270799	911.0	1.124144	0.032944
3	1.132960e+06	-644298.185858	209.978958	6.146258	4.459189	-4.719252	1666.0	1.055472	-0.023361
4	1.133760e+06	-644598.185858	185.525940	34.531841	181.556056	27.067570	309.0	3.279520	-5.745003

Tests

• Added a test to ensure that NaN padding is applied if a chunk fails.
• Added a test to verify the correct behavior of BlockwiseCoreg.stats when all chunks are processed successfully.
• Modified the test_blockwise_coreg_large_gaps test to ensure that the len of stats is the same than the number of subdivision (i.e. each chunk has stats, even if the chunk fails -> NaN stats).

Documentation

Updated the docstring for BlockwiseCoreg.stats to reflect the NaN padding when a chunck fails.

[quinto-clara]

Two requests :

• The index used for the loop that build statistics needs to be changed
• The center_x, center_y and center_z values needs to be kept when a chunk fails

[quinto-clara]

The BlockwiseCoreg method handles now missing data, improving robustness of the stats method.

The changes regarding the the two previous requests are approved :

• The index used for the loop that build statistics is now correct. All the chunks, even failed ones are taken into account.

• The center_x and center_y values are now kept when a chunk fails. Thus we can show where the NaN values are :

Example with 64 subdivisions and an altered DEM :

Example with 65 subdivisions (no altered DEM). We can visualize which chunk failed :

[rhugonnet]

Great!
I have nothing to comment on, all good for me 🙂 (I assume the disabling of license headers in the last commit is temporary?)

Tagging @erikmannerfelt (who largely coded BlockwiseCoreg) so that he sees this when he has the time.
He is likely not available for the rest of the month, so don't wait on it for merging 😉

[adebardo]

Thanks @rhugonnet, Yes, it’s temporary, just until we find a solution to avoid putting the CNES copyright on code it didn’t develop.