search

torrvision / spaint

A framework for interactive, real-time 3D scene segmentation
http://www.robots.ox.ac.uk/~tvg/projects/SemanticPaint/index.php
Other
192 stars 53 forks source link

How to improve the accuracy of reconstruction #10

Closed 627633854 closed 6 years ago

627633854 commented 6 years ago

When I scanned a complex model, I could see the details of the model in the spaint window, but the accuracy of the model displayed after the reconstruction was much worse. Is there any way to improve it?

sgolodetz commented 6 years ago

Change the voxel size down to something sensible in spaintgui's main.cpp possibly - do you have a screenshot?

627633854 commented 6 years ago

yes, i scan a Hulk model, but lost a lot of details. image
and the result image

sgolodetz commented 6 years ago

What's the scale of the model in comparison to the voxel size?

627633854 commented 6 years ago

The size of model is 25cm,21cm,13cm，is it small?

sgolodetz commented 6 years ago

Yes, that's pretty small - if you're using 0.5cm voxels, that means your model is only 50 voxels across. Plus, there's depth accuracy to consider - at that scale, the fine details in the model are unlikely to be captured well by a standard depth camera, so it's not just a matter of making the voxel size smaller (although you could try that, to get some improvement).

627633854 commented 6 years ago

OK, thanks a lot