Support Website Contact Support Blog

"Fuzzy" Edges

I’m (almost constantly) getting a lot of noise around the edges of an object I’m trying to model.

Do I need to manually delete these points?

Hello,

I’d be really interested to look at your project. Do you mind sharing it?

Thanks in advance.

Sure… um, how do I do that? :slight_smile:

You can share your project through a file sharing service (Google Drive, Dropbox, WeTransfer,…)

Gotcha - wasn’t sure if Pix4D had a unique uploading space. Stay tuned…

Hello Bob!

The best practice for modeling the objects is to annotate images by selecting the unwanted content on the images just after step 1. You can learn more how to Annotate Images in the rayCloud here

The annotation tool will let you mark areas on the photographs which in your opinion are redundant and may cause the model to be damaged by adding noise in further steps of processing.

It is a perfect tool which will give you the ability to act proactively.

Explore and have fun! :slight_smile:

Try these links:

Example 1 (from above):

https://drive.google.com/open?id=1P5p_DP9Ry6Ee-KPXQkoQPaJT6L6VdPEg

 

Example 2 (similar object, different camera and environment):

https://drive.google.com/open?id=1_SFDu7y0qz4esX0T2wKKKpy8IvaqFzuU

 

Thanks, Beata - I’ll give the masking/carving a try.

Thanks a lot !

Those data will help us to work on our algorithms !

 

Have a nice day.

Here’s my two cents

  1. Get closer to object. You are taking pictures too far away.

  2. When you can control the environment take the pictures in a place where the background isn’t  matching color to the object (makes carving alot easier)

  3. More overlap so you can calculate point cloud with “Minium number of matches to 5 or 6” that way you’ll get much cleaner point cloud

Here’s my 3D-model: 6 carved pictures and a little manual point cloud editing. Took like 5 minutes. Still room for an improvement

 

I was about 24 inches away, but with a wider angle lens. If I get right on top of the subject, is there any problem with not having enough background for parallax to aid the algorithm?

The carving seems to be a big help. I thought the carving would be retained in the project but it’s all wiped out if you re-process step 2 (still trying to learn what processing wipes out what stuff and what the difference is between the “reoptimize” and “rematch + reoptimize” functions).

I realize there are “help articles for all these things” but they’re not very helpful. I’d love to have a technical document about what the actual functions are that are being performed rather than “This process reoptimizes the internal and external camera parameters.” It’s just me having to learn the application.

I also don’t have access to a studio where I can paint the walls green or some contrasting color - the Pix4D “Eagle” project is a good example of low-contrast modeling and seems to work well.

https://support.pix4d.com/hc/en-us/articles/360000235126-Example-projects#label5

As far as overlap, I was taking the shots every 15 degrees at two different hieghts - this should be enough (again, referring to the Eagle sculpture, they used 44 pics and the overlap is about what I have (actually a little less) and that model works very well.)

This is one of my tests

161 pictures, 3 orbits around the target and few pictures straight from above

EOS 6D with 17-40L at 17mm f8.0 as close as possible to the walker. I used tripod and remote trigger

These settings:

This is the result without any editing to point cloud: