Multispectral alignment issues in v2.3.1

[mbohman]

I have been encountering errors in recent uses of ODM for processing multispectral imagery. ODM appears to be matching one band, as per the new parameter primary band, but then subsequently fails to match or add any other band to the reconstruction and final output. I have included a snippet of the relevant error message.

INFO] Computing band alignment
[WARNING] Compute homography: OpenCV(4.5.0) /code/SuperBuild/src/opencv/modules/calib3d/src/fundam.cpp:385: error: (-28:Unknown error code -28) The input arrays should have at least 4 corresponding point sets to calculate Homography in function ‘findHomography’
[INFO] IMG_0021_1.tif → IMG_0021_2.tif cannot be matched

I have encountered this error while using datasets which were processing fine in previous versions of ODM. I have tried using small datasets(such as the 90 image folder I am sharing below) and larger datasets of 3000, and this error was persisted. Additionally this error does not seem to be correlated with any parameter that I can find, including changing the primary band which is used.

I am including a link to a google drive folder containing 90 tiffs, along with the full log and output folder from running them in v2.3.1 Any insight into this issue would be greatly appreciated.

https://drive.google.com/drive/folders/1X1puWUz2mCseaLOJnOxrA-NWGzUFwgfX?usp=sharing

[pierotofy]

Thanks for reporting the issue (and providing some test data)!

I made some improvements in https://github.com/OpenDroneMap/ODM/pull/1221 and was able to process the dataset with these changes.

We’ll release shortly.

[pierotofy]

[pierotofy]

[mbohman]

It’s working great on my end following your update. Much faster and higher quality than previous versions, thanks for all your work.

[mbohman]

I’ve been able to reproduce your results with images sets from 100 - 500 images. However I’ve attempted a run with 3000 images and received incorrect results. Below is the orthophoto from one such run.

I’m also linking a google drive folder with the 3000 tiffs I used to get the above ortho, the full log and output files. Its uploading now and should be fully there in an hour or so. I’m also trying a few different flights, along with a few different settings to see if I can achieve a good result.

https://drive.google.com/drive/folders/1Jo2WX4acGIEdP0D-GdRruFTJDz7ykKDJ?usp=sharing

[mbohman]

I’ve been able to achieve good results on a different flight’s dataset. I’m not sure whats different about the above dataset, but it’s been able to be processed on previous versions

[pierotofy]

Look up “bowling” on the forum. Also https://docs.opendronemap.org/tutorials.html#calibrating-the-camera