Point Cloud Decimation

It was discussed before and actually mentioned that point cloud decimation was reduced and that we would be receiving more points. This model in particular only has 12.8m points whereas Pix4d, Agisoft Metashape and Simactive Correlator all have 14m, 16m, and 17m respectively. It is obviously by DroneDeploy’s point cloud that breaklines we not held and a grid method was applied. We need to fix this because a 3ft cube should be easily recognizable…

@Jamespipe @Andrew_Fraser

Michael have you ever processed a full map with Max settings on a personal PC (Not Cloud) with Pix4D? The point clouds are much more dense. Unfortunately drone point clouds have a long way to come before they will be able to be used easily on Civil3D. I’m excited to try the scan feature with one of my Skydio’s when they push the alpha to me.

We actually use Simactive Correlator. It has a much better bundle adjustment and DTM processing. Point clouds are usually 4x the size of DroneDeploy’s. We then use Carlson Precision 3D Topo to edit the point clouds before making surfaces and before putting them in Navisworks.

@MichaelL Was this processed in Structures mode, or Terrain mode?

This one was terrain, but the fact remains that there are 30% less points than a half-size image medium processing in Correlator which are their “Terrain” settings that process in about 3 hours locally.

Even though, here is a structure mode that exhibits the same poor quality due to missing points.

We’ve also got a pretty evident problem with noise in Structure mode which makes it confusing why it removes so many vertical objects.

My opinion is that the decimation needs to be stated instead of telling us that it is a “Max Points” file. The gridding is an obvious sign that this is at least a 1/10 decimation.

*I am going to reprocess the project from the first image with Structure mode now. I am sure it will be better, but the Terrain mode should have been better than it was. The amount of space that is skipped is not going to catch many of the types of slopes that are on Civil sites which makes 2-3cm accuracy unobtainable even with PPK.

So this is what I got when trying to process nadirs with Structure mode. This has never happened before. Not good.


1 Like

I have been getting this more frequently than not. SUPER FRUSTRATING after I spend an hour tagging GCP’s.

Casualties of flying too low and trying to use GCP’s in Structure mode processing.

Have you guys reached out to our support team? I recommend doing so any time you run into unexpected results. We can help pinpoint why a result turned out a specific way, and offer steps to avoid it next time. Additionally, if greater PC density is needed, I recommend letting your account team know, as we have the ability to create more dense point clouds if you need.

In our case it’s not isolated and the Team has known for quite some time. It wouldn’t be a special order situation either. All clouds that we download should have more points. Can you find out what factor is being used?

Another effect of decimation. We had this problem a while back and the support team was able to resolve it in reprocessing somehow. Well, it’s back. The is the second job I have seen it on in the last week. Both subject were slightly homogenous in the missed areas so I understand a little difficulty in processing, but there are plenty of tie-points visible which has been verified by other software. Coverage was good according to the DroneDeploy processing report.


(As Exported)

@Andrew_Fraser, @Jamespipe

If I zoom in I can start to see it in DroneDeploy. Maybe displayed pixel size is different between softwares…

Hi Michael,

Please raise this via support - the computer vision team will want to investigate. This appears unrelated to previous issues reported on this thread, and not due to any kind of deliberate decimation.


1 Like

Will do, thanks James!