Failed to create orthophoto

#1

Hello, guys! I use ODM to create orthophoto from a set of data sets. But the result is so bad, and the coordinates of point could(reconstruction.ply) is error. The overall orientation is the opposite in reconstruction.ply! I use openMVG to successfully calculate point clouds, and the direction of the point cloud is normal.
reconstruction.ply:
pc
the orthophoto:
odm_orthophoto
openMVG:
mvg
Who can use this data sets(the ccd sensor size is 12.833, --force-ccd 12.8333) to test ODM against? Help me, thank you!

#2

Works fine for me with the default Hi-Res settings: https://dashboard.aerosurvey.co.nz/public.cgi?S=326&T=b3eb358b-669e-4940-90fa-516fb7d0f7e7

What options are you using?

#3

Hello, I used the default setting. And I can not open the link: https://dashboard.aerosurvey.co.nz/public.cgi?S=326&T=b3eb358b-669e-4940-90fa-516fb7d0f7e7
Can you use google drive? Thank you!

#4

You are in a region that was blocked at the firewall (China). Try now.

Maybe you should re-run the task and see if the results come out differently. I used the regular High Resolution setting.

#5

The link also cannot be opened via VPN.
I found that the point cloud coordinates generated by openSFM are wrong.
Do you use WebODM or NodeODM?
I directly use: ./run.sh images --force-ccd 12.8333

#6

Can you verify that you have exif data for GPS location in the data, and that the exif data is correct?

#7

Hello, I am very sure that there are exif data for GPS location and the exif data is correct. You can try this dataset.
I tried a variety of different data sets and debug. I found a very strange problem. When I only use some images that their GPS position is growing linearly(like picture A), the orientation of the point cloud calculated by OpenSfM or the dense point cloud calculated by mve is wrong. Then the 2.5D mesh calculated from the point cloud or the dense point cloud is wrong! Finally, this results in the inability to produce the correct orthophoto.
When I add more images to the original, some GPS coordinate values of images are reduced in return flight, like Picture B. The orientation of the point cloud is right, then the 2.5D mesh is right, and the it can product the correct orthophoto.
I don’t know the reason between the two results. But this situation does exist! I don’t understand why OpenSfM uses local coordinates. This may be a bug that lead to the above phenomenon. Because I use openMVG to successfully calculate point clouds, and the direction of the point cloud is right! You can try it.
I hope to attract the attention of the ODM author and you. Thank you!
Picture A:
g3
r1

Picture B:
g2
r2