Hello,
I tried other times to reconstruct objects by taking photographs by circling the object and then placing the object on different sides, repeating a series of other shots around the object. I got very good results. However, there are circumstances where it is difficult to rotate around the object. So these days, I have been trying with a turntable. I placed a small “THUN” teddy bear by making five rings:
The first loop was in a sitting position, and I rotated the platter in steps of about 10 ° at a time.
I took the photos in the second round with the teddy bear lying on his back. Again, by rotating the rotator about 10 ° between one click and the next.
Third loop, I took the shots by putting him on his stomach.
I took the photos in the fourth lap putting the teddy bear on the right side.
I took the photos in the fifth lap putting the teddy bear on the left side.
The result was very bad (all camera poses have been correctly registered). As if the reconstructions were intertwined. The turntable (its plot) was perfectly rebuilt. I have made other attempts, trying with handmade masks (simulating a green screen), but I have not obtained any results.
Maybe I should use export n components for n loops and then merge? So I wonder then the components would realign correctly?
Could you help me figure out the correct workflow with a turntable in Reality Capture?
Thanks for your reply and suggestion.
Certainly, the result has improved enormously. But it still doesn’t work. The individual loops are perfectly aligned. I then cut the cloud appropriately and generated the mesh and masks. Obtained all the masks, I imported all the photos with the relative masks, but I got 2 separated components:
One with 44 cameras registered on the 205 imported and one with 39 cameras on the 205. Then I have a folder with Small Components (16 with less than 21 cameras).
Why is this happening? In the past I had obtained an excellent result by photographing different sides of an object, but rotating around the object. I would like to use turntables because it makes the procedure much faster in some contexts.
I have attached some screenshots:
1DS:
Model component 16:
Model component 17:
As you can see, the first model lies on the right side, the other on the left side. So why did RC create separate components? Yet, the reconstructions of the individual loops were correct and without triangles outside the reconstruction box.
As you can see, the edges and the object’s surface are jagged. In reality, even the mask photos are very irregular in their outlines.
I would be grateful for any help and consideration on this result.
Hi, how many images did you take in each loop? As I can see, you have a lot of components with a small amount of images (3 - 14). Were you able to align circles around your object separately to create masks without any problems? This seems like there is not enough overlap between images.
Did you create masks from a Preview model? How do they look?
Hi ,
Thank you for your consideration.
In the first loop, I made 31 shots, in the second 38 shots, in the third 38 shots, in the fourth 44 shots and in the last 54 shots.
In the alignment of the individual loops to make the masks, I had 100% of the alignment of the shots. I made the masks from Normal model quality.
Could the problem be the texture of the object with few details? However, I do not understand why individual loops had 100% cameras aligned in the alignment phase.
I will try to use the control points. I will update you as soon as I try the procedure. Do I apply the control points separately to each loop from which I will get the various masks?
Thank you so much for your support. Below I’ll show you a test I did very quickly around a small amphora and no turntable but turning it at each loop on a different side. The result is excellent considering that I took it really fast and with a smartphone.
However, I am sorry about this last result, also because I would like to achieve the proper technique to use a workflow with a turntable. For example, would it be helpful to use a background similar to a green screen?
I forgot, for this project my alignment parameters are as follows:
yes for a turntable the background needs to have no trackable features, normally completely black or completely white. If RC matches features in the background, these features are telling the software that the camera is stationary. The idea of a turntable is to trick the software that the camera is moving around the object, so any data suggesting the camera is stationary is going to cause you problems.