Support Website Contact Support Blog

Advice sought on poor model

Looking for some guidance here. I created a 3D model from around 290 photos of the portico at University College London (I’m working with their Geomatics department). This was done with a Phantom4, using mostly obliques, but also some Nadir. All done in free flight, but with plenty of overlap. After the initial run of steps 1 and 2 the model came out as shown below:

Putting aside the issues of sky, this looks very poor in my opinion. For example, look at the steps in front of the portico.


I then painstakingly did the annotation to all the photos with sky showing and re-ran step 2 but this appears to have introduced some further issues, as shown in the following 2 screen grabs. For example, the rear of the domed roof is very disappointing (far worse than before I did the annotation).


By comparison, I ran the same photos through Drone Deploy (the free version) and got a really good model and did not even have to bother with annotations:


So, my question is, what am I doing wrong?

What are your project processing specs?  

It would help if we could see the position of the images in the project (e.g. a screenshot in the rayCloud view with the cameras activated) and the quality report. In the meantime, here are a few tips that could help for next time: 

I’m not sure which annotation tool you have used. I would recommend to use the Carve tool after step 2 is processed. In this way, you can directly see the effect of the annotation on the densified point cloud and you do not have to do guess work. This will probably also reduce the amount of images that you have to annotate to achieve a good result. The 3d textured mesh can be regenerated with Process > Generate 3D Textured Mesh once the point cloud looks ok. More about cleaning the sky with the Carve tool in this article: [

]( the point cloud editing tool can be useful as well:

The effect of the sky can be reduced a lot by improving the image acquisition. If possible, the camera should point downwards (also when oblique) so that the sky is not present in the images. For a 3D reconstruction such as this one, I’d suggest the double grid flight plan in Pix4Dcapture. This will make sure the images are taken with enough overlap and in a consistent manner. 

For areas such as the stairs, it might help to add Manual Tie Points (MTPs) and to Process > Reoptimize. Note that this should be done before processing step 2, so that you do not have to regenerate the results of step 2: 

If you could post a link to the images of the project and if possible to the quality report, we could have a look to see if something else can be improved.