parallel_stereo completes without generating PC.tif

[forwork]

Hello,

I am running parallel_stereo with ISIS cube inputs and adjusted camera from a bundle adjustment. The pipeline completes its initial steps, but it exits without generating essential downstream files like D.tif, RD.tif, and PC.tif.

My Exact Command:

parallel_stereo \ --stereo-algorithm asp_mgm \ --clean-match-files-prefix ba/run \ img1.cub img2.cub \ ba/img1-adjusted-state.json img2-adjusted-state.json \ stereo/run

Context & Observations:

• Data: I am using full strip cube or cropped cube from Chandrayaan-2 OHRC.

• There are no crash error while parallel_stereo is running, but the execution succesfully completes without an complete set of output files.
  

• Everytime after Correlation completes there is a warning for lowering memory consumption.

• I tried using entry and exit points to force the pipeline to make PC.tif. However, it stops because files from the previous step are missing giving the file not found error (ex: RD.tif not found, D.tif not found), so it cannot proceed to the next step.
  

• The run-log-stereo-corr log says full-res search range based on D_sub.

I wanted to ask that under what circumstances or what causes the PC.tif to not be generated and how can be resolve this issues as PC.tif will be need in further point2dem command.

Thank you so much.

[Oleg Alexandrov]

All these things you see by themselves are likely not an error.

I did fix an error with empty PC.tif for Chandraayan (I think it was for TMC though), that requires a very recent build.

One should ensure the build is installed like this: https://stereopipeline.readthedocs.io/en/latest/examples/chandrayaan2.html#environment-setup

Or even better, use our just-released version 3.7.0:  https://stereopipeline.readthedocs.io/en/latest/installation.html#conda-based-installation

More general directions are below if that does not help.

The first thing to look at is in the stereo_pprc log for the stereo convergence angle. If it is tiny, say under 1 degree, the rays from the cameras are almost parallel and no point cloud will exist.

It is suggested to do gdalinfo -stats on the intermediate produts, which are F.tif, D.tif, D_sub.tif. if a lot of invalid pixels are shown so the statistics of the third band are almost all 0, that means the input dataset is not good.

RD.tif not found, etc, is a big problem early on. I guess if you use the entry point while not having the data for that entry point it won't work.

Look at L.tif and R.tif and see how they appear.

Can also take a look at the match file plotted against the images it was made with. In stereo_gui documentation it is shown how to plot it.

Try to see if the provided example works for you.

Try to look at the input files you pass to it in stereo_gui.

So, likely your pair is not good, or something is not right with the procedure. 

If no luck at all, let me know. 

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/a0a1faa7-300b-4d3c-8054-055552b1602bn%40googlegroups.com.