Okay, let me explain it to you.
When you scan markers for me, each frame aligns based on the markers. This means that features are ignored, and only the markers are used to align your frames. This saves computing time, and features cannot cause confusion.
Here you can see, for example, the manual marker detection by a Discord user with an ML model (no idea how Revoscnaa does it).
This is from Revoscan exported:
The problem with this is that the scanner does not have 100% accuracy, and with each frame, the frames can shift. Imagine you have a puzzle with pieces that don’t fit 100%. You place the first piece, and the accuracy is 99.998%. Now you place another one, and the accuracy error adds up…
This is what it looks like with the markers:
The optimization takes all frames (aligned by the markers), ignores the markers, and only uses the features of the object to refine the alignment further. But as mentioned, this is based on the object, similar to CloudCompare – I suspect also using the ICP method. Your markers may appear blurred afterward, but the object itself will have better accuracy, which is great!
Watch the video and pay close attention at 0:07 to how the markers suddenly shift, but the lion’s paw improves.
You can also see on the construction that the hole suddenly disappears because the raw material overlaps better, and the RevoScan algorithm doesn’t have to remove anything (noise filter).
The biggest effect can be seen when the markers are not placed directly on the object but next to it. When I place the markers directly on the object, the fusion process makes it worse.
Fun fact:
In turntable mode with the Metrox, multiple frames are captured at the same position. The frames are then compared, and only the parts with overlap are kept, effectively removing noise. What we see in RevoScan is not the raw material – there are already several algorithms in place that filter out noise and other artifacts.
Credits goes out to @trg_ldn_22573 on Discord!
Hes doing a great Job and @Revopoint-Jane i think he really can help to improve RevoScan!
In my opinion, he should get free access to the SDK so he can work more with the data, Revopoint would benefit greatly from his suggestions.