Good or Bad dataset

#1

From @PaoloCaf on Wed May 18 2016 13:12:30 GMT+0000 (UTC)

Hi,

I’m trying to use opendronemap with my dataset.
For some reason the final point cloud is empty and the reconstruction doesn’t complete.

I get the following error:

running “/home/ubuntu/OpenDroneMap/bin/odm_meshing” -inputFile /home/ubuntu/surv ey/reconstruction-with-image-size-2400-results/option-0000.ply -outputFile /home /ubuntu/survey/reconstruction-with-image-size-2400-results/odm_mesh-0000.ply -lo gFile /home/ubuntu/survey/reconstruction-with-image-size-2400/odm_meshing/odm_me shing_log.txt -maxVertexCount 100000 -octreeDepth 9 -samplesPerNode 1.0 -solverD ivide 9
[pcl::KdTreeFLANN::setInputCloud] Cannot create a KDTree with an empty input clo ud!
[WARNING] Initial residual too low: 0 < 0.000001
[WARNING] Initial residual too low: 0 < 0.000001
[WARNING] Initial residual too low: 0 < 0.000001
[WARNING] Initial residual too low: 0 < 0.000001
[WARNING] Initial residual too low: 0 < 0.000001
[pcl::io::savePLYFile] Input point cloud has no data!
Error in OdmMeshing:
Error when saving mesh to file:
/home/ubuntu/survey/reconstruction-with-image-size-2400-results/odm_mesh-0000.pl

The pmvs didn’t report any issue but the option-0000.py is empty.

How can I troubleshoot this problem and find if there is a problem on my dataset?

thanks,

Paolo


Copied from original issue: https://github.com/OpenDroneMap/OpenDroneMap/issues/310

#2

From @dakotabenjamin on Wed May 18 2016 15:13:03 GMT+0000 (UTC)

How many photos are in your dataset? Can you post a sample photo? Can you post the stdout of the following call:

ls -l /home/ubuntu/survey/reconstruction-with-image-size-2400/opensfm/
#3

From @PaoloCaf on Wed May 18 2016 17:12:44 GMT+0000 (UTC)

Give me one day.
I did it in aws and terminated the instance.
Il 18/Mag/2016 17:13, “Dakota Benjamin” notifications@github.com ha
scritto:

How many photos are in your dataset? Can you post a sample photo? Can you
post the stdout of the following call:

ls -l /home/ubuntu/survey/reconstruction-with-image-size-2400/opensfm/

—
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub
https://github.com/OpenDroneMap/OpenDroneMap/issues/310#issuecomment-220058560

#4

From @PaoloCaf on Thu May 19 2016 12:41:55 GMT+0000 (UTC)

Ok … ready.
I have 51 images.
This is a sample image:

https://dl.dropboxusercontent.com/s/e0t9dzyf5d6jf2x/IMG_0879_geotag.JPG?dl=0(image larger than 3072KB)

Images have been already renamed with small characters.

ubuntu@ip-172-31-31-46:/mnt/OpenDroneMap$ ls -l /mnt/survey/reconstruction-with-image-size-2400/opensfm/
total 125808
-rw-r–r-- 1 root root 7082298 May 19 12:34 bundle_r000.out
-rw-r–r-- 1 root root 336 May 19 12:19 camera_models.json
-rw-r–r-- 1 root root 108 May 19 12:19 config.yaml
drwxr-xr-x 2 root root 4096 May 19 12:19 exif
-rw-r–r-- 1 root root 1173 May 19 12:19 image_list.txt
-rw-r–r-- 1 root root 1019 May 19 12:34 list_r000.out
drwxr-xr-x 2 root root 4096 May 19 12:29 matches
-rw-r–r-- 1 root root 138 May 19 12:34 profile.log
-rw-r–r-- 1 root root 19333944 May 19 12:34 reconstruction.json
-rw-r–r-- 1 root root 90482682 May 19 12:34 reconstruction.meshed.json
-rw-r–r-- 1 root root 79 May 19 12:29 reference_lla.json
drwxr-xr-x 2 root root 12288 May 19 12:20 root_hahog
-rw-r–r-- 1 root root 11723644 May 19 12:29 tracks.csv

I will keep the EC2 instance available for the next 2 hours more tests are needed.

Paolo

#5

From @dakotabenjamin on Tue May 31 2016 13:24:48 GMT+0000 (UTC)

1. you may have too few images
2. your overlap may be insufficient, leading to a sparse point cloud that cannot be densified in PMVS.

#6

From @dakotabenjamin on Wed Jun 15 2016 14:55:31 GMT+0000 (UTC)

Any luck @PaoloCaf?

#7

From @PaoloCaf on Wed Jun 15 2016 15:17:40 GMT+0000 (UTC)

Hi,

Unfortunately I had no chance to test it again due to a crash and then bad
weather.
This weekend the forecast says that it will be sunny and I will most
probably fly again and let you know.

thanks a lot.
Paolo

2016-06-15 16:55 GMT+02:00 Dakota Benjamin notifications@github.com:

Any luck @PaoloCaf https://github.com/PaoloCaf?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/OpenDroneMap/OpenDroneMap/issues/310#issuecomment-226213226,
or mute the thread
https://github.com/notifications/unsubscribe/ASh7eAEPpQcSd-ckI1Gfe9Finh0fwCu2ks5qMBJogaJpZM4IhRCK
.

#8

From @rionlerm on Fri Jul 08 2016 12:18:35 GMT+0000 (UTC)

Hi @PaoloCaf and others,

I recently found sidelap below 60% and overlap below 80% does not produce results in ODM :frowning: Also, here are some LAS file results (converted to Digital Surface Model) on top of the orthophoto created by OpenDroneMap (ODM). As you can see the dotted black line shows only two transect lines that were flown at 120 metersATL. This was not adequate for the proper creation of an elevation file in ODM.

inadequatetransectlines

#9

From @dakotabenjamin on Fri Jul 08 2016 16:08:29 GMT+0000 (UTC)

Great work @rionlerm this is good information to know. When the side/overlap is too low, where does ODM fail?

#10

From @smathermather on Mon Jul 11 2016 20:03:23 GMT+0000 (UTC)

Yes @rionlerm, it would be good to know if it fails at sparse point cloud creation or dense (or somewhere else).

#11

From @rionlerm on Thu Aug 04 2016 06:27:21 GMT+0000 (UTC)

Hi @dakotabenjamin and @smathermather ,

I apologise for never replying to your last questions.

I accidentally calculated transect width incorrectly with a flight and resulted in prob 40% overlap and 30% sidelap. ODM produced the following folders and contents only:

.
├── images
├── images_resize
├── opensfm
│ ├── exif
│ ├── matches
│ └── root_hahog
└── pmvs
└── recon0
├── models
├── txt
└── visualize

I think the script stalled at Expand patches...

Please mention if you need to see any additional files in these directories.

#12

From @dakotabenjamin on Thu Aug 04 2016 14:33:27 GMT+0000 (UTC)

Anything under 60% is going to be hard to reconstruct. You can try using the parameter --pmvs-minImageNum 2 which means that PMVS will only require 2 matching images to reconstruct rather than 3.

Otherwise, you may be able to play around with other cmvs-pmvs parameters to get it to run without crashing. This is a resource I use often to try to get things working the way I want.

#13

From @rionlerm on Fri Aug 05 2016 06:17:42 GMT+0000 (UTC)

Thanks Dakota!