Artifacts on slopes

[Rachael Hoover]

Hello! I am wondering if you have any suggestions for dealing with holes and artifacts on slopes. I have tried adjusting the search radius factor while using point2dem and that doesn't solve the issue. I am wondering if attempting dem_mosaic is the best path forward? My concern with that would be that I may fill the holes with this method but I would probably also be amplifying the artifacts. I am attaching a figure showing my issues. I am mainly concerned with the western facing slopes. 

Thanks in advance for the help!

Rachael

[Alexandrov, Oleg (ARC-TI)[KBR Wyle Services, LLC]]

Steep slopes are tricky to resolve, because they may differ a lot in left and right images. More precisely, sometimes one image shows them well, but the slopes can either be occluded or "shrank" in a second image. 

You can try to see if --stereo-algorithm 2 works better for you (with its default subpixel mode or with --subpixel-mode 3).  This can be run on a small clip in stereo_gui for experimentation. 

Our doc also talks about how one can get an initial ASP DEM, imperfect as it is,  as what you are getting, then create a low-resolution DEM by going back to the point cloud and running point2dem with a grid size larger by a factor of 10 maybe, mapprojecting onto this, then re-running stereo with the mapprojected images.  

This will make the left and right images be more similar where the slope is, hopefully, and will resolve them better in stereo.

Oleg

---

From: ames-stereo-pi...@googlegroups.com <ames-stereo-pi...@googlegroups.com> on behalf of Rachael Hoover <hoov...@gmail.com>
Sent: Tuesday, June 1, 2021 11:37 AM
To: Ames Stereo Pipeline Support <ames-stereo-pi...@googlegroups.com>
Subject: [EXTERNAL] Artifacts on slopes

 

Hello! I am wondering if you have any suggestions for dealing with holes and artifacts on slopes. I have tried adjusting the search radius factor while using point2dem and that doesn't solve the issue. I am wondering if attempting dem_mosaic is the best path forward? My concern with that would be that I may fill the holes with this method but I would probably also be amplifying the artifacts. I am attaching a figure showing my issues. I am mainly concerned with the western facing slopes. 

Thanks in advance for the help!

Rachael

--
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 on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/88478417-5d45-4781-9fd5-548b8d240798n%40googlegroups.com.

[David Shean]

Hi Rachel,

Building on Oleg’s thoughts - would be useful to see the left and right images you’re passing to ASP to assess illumination and view angle considerations.  There’s only so much you can do if your second image doesn’t capture the slope well.

Do panels A and B have identical extent?  Is the position of your yellow ellipse identical in A/B, relative to ground features?

I’m wondering if the artifacts are actually concentrated along the slope break at the base of the slope, rather than the steep slope face.  At the base of the slope, you have a linear feature with brightness difference due to illumination, which can cause problems for the correlator, leading to artifacts.

-David

To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/MN2PR09MB54498E5667521FEAD9DE240C883E9%40MN2PR09MB5449.namprd09.prod.outlook.com.

[Rachael Hoover]

Hi David -

I added a semi-transparent DEM over the image and I think the artifacts are located mostly within the middle of the slope. 

For the left image I used P06_003489_2176_XI_37N130W (Ls 225.7, Incidence 72.98 and emission 2.55) and the right image I used P02_001656_2176_XI_37N130W (Ls 144.7, incidence 50.97 and emission 0.83)

Rachael

[Beyer, Ross A. (ARC-SST)[SETI INSTITUTE]]

Rachael,

Are the slope streaks on those slopes different between the two images?

If the slope streaks or deposits at the foot of the slopes are different between the two images, you may need to NULL out the streaks so that ASP doesn’t try and correlate them. The correlates do fine with two pictures of a horse, but if someone spray-paints the horse to look like a zebra between picture1 and picture2, even though the horse has the same 3D shape, the image correlator will have problems. However, I’d leave this for a last resort.

If so, that might complicate the matcher.

Are you doing bundle adjustment prior to stereo? If not, that might help, too.

Definitely follow Oleg’s advice, and play with a small cropped area to make experiments go faster. The SGM (--stereo-algorithm 2) or MGM (--stereo-algorithm 3) algorithms might do better, as he suggests, and playing with --subpixel-mode.

Additionally, following an iterative process to slowly build up to the DEM resolution you want (rather than trying to go for the full-resolution first off), following https://github.com/USGS-Astrogeology/asp_scripts (and LPSC abstracts referenced therein) or https://github.com/AndrewAnnex/asap_stereo might help.

> --
> 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 on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/7527ed3a-4a0f-4eaa-ac67-7e098f5a9309n%40googlegroups.com.
> <P06_left.png><DEMOverlay.png><nooverlay.png><P02_right.png>

Ross

http://RossBeyer.net/science/

[Rachael Hoover]

Hi Ross - The slope streaks are different between images. If I end up needed to, how would I NULL out the slope streaks? 

I will go ahead and follow Oleg's advice and look into a more iterative process also.

Thanks everyone for the help!

Rachael

[Beyer, Ross A. (ARC-SST)[SETI INSTITUTE]]

Rachel,

> Hi Ross - The slope streaks are different between images. If I end up needed to, how would I NULL out the slope streaks?

Take your HiRISE image(s), make copies of them, and then you can use the ISIS qview program to edit those image copies. In the qview window, there are icons along the right-hand side, and you can select the pencil Icon (Image Edit), and the tool context menu will allow you to convert individual pixels, lines, or rectangles of pixels to the ISIS NULL value.

> I will go ahead and follow Oleg's advice and look into a more iterative process also.

Hopefully that will be more fruitful. Good luck!

> To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/703de75a-6b8e-4994-84e4-8c6d5fcbef4bn%40googlegroups.com.

Ross

http://RossBeyer.net/science/

[Rachael Hoover]

Hi All - 

As a follow up to this - I have had much better luck with the SGM and have made better DTMs with less artifacts on the slope for one area (yay!). Now I am trying the SGM on a different set of  images and keep running into an error for Correlation - Step 1....

"Warning: Unable to compute valid search ranges for SGM input!."

I have tried adjusting the the --corr-tile-size parameter up to 9000,  have tried using the default value, and have also tried this with parallel_stereo (I didn't get the same error but it output blank point clouds). 

 Any other ideas?

Thanks!

Rachael

[Alexandrov, Oleg (ARC-TI)[KBR Wyle Services, LLC]]

You should use parallel_stereo instead of stereo, and MGM (--stereo-algorithm 2) works better than SGM (--stereo-algorithm 1). Then the --corr-tile-size parameter need not be set.

I am very confident that this should work in general, but there may be quirks due to your particular data. 

Since I recall you use HiRISE, you can try starting stereo_gui with exactly the same options you intend to pass to parallel_stereo (including the output prefix), zoom in, and choose some "easy" areas first with Control-Mouse-Drag in the left and right images, and run parallel_stereo for the menu. 

Hopefully your point cloud will not be empty. If still having issues, you can load in stereo_gui your resulting .match file and corresponding images and see if something went wrong there (the stereo_gui page has more info on how to load a match file).

Then, if confident enough, you can run bigger clips or the whole image set. 

Oh, and usually stereo does the worst when the left and right images are visually very different, either due to perspective or illumination. 

---

From: ames-stereo-pi...@googlegroups.com <ames-stereo-pi...@googlegroups.com> on behalf of Rachael Hoover <hoov...@gmail.com>

Sent: Tuesday, August 31, 2021 12:49 PM

To: Ames Stereo Pipeline Support <ames-stereo-pi...@googlegroups.com>

Subject: Re: [EXTERNAL] Artifacts on slopes

 

To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/7cdfba80-f8fe-486c-9b45-54193a7550c8n%40googlegroups.com.

[Scott McMichael]

If you still need to debug what is going on with MGM you can try increasing the stereo log level (see section 2.3 in the manual) to get more information.  Increasing the tile size in MGM usually leads to memory problems but I can look at the log output to try and figure out what is going wrong.  If it is working in a small test image then there may be a particular region of the image that it is having trouble with and you may need to experiment with that region separately.

Scott

To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/MN2PR09MB54493324CAFA014D6420C82488CC9%40MN2PR09MB5449.namprd09.prod.outlook.com.

[Rachael Hoover]

running MGM worked great. Thanks!