Texture reprojection problem : near pixel bleeding

Hi,

We shooted a bronze head on a wood support. It perfectly aligned, it perfectly reconstruct, we clean it removing the support and texturize. Some shot have been taken by underneath the statue so the support intersect the vue with some lower part of the statue. When we texturize, we had a brown bleeding of the support color projected on bottom part of the statue.

So we tried different method to avoid this problem (It’s an history of art project on 5 year on museum pieces and the goal is to present the best acquisition methodology without cheating on data … so no use of PS to correct smthg, we will work on the dense point cloud for annotate micro traces by segmentation). :

  • deactivate the pictures that project this bleeding : it works but as we remove 120 pictures we have lost details on some parts of the texture … not possible.

  • Reduce the weight of those pictures : still bleeding but lighter and lack of details such in the first try.

  • Masking only the statue and reconstruct and texturize : still reconstruct parts of the support and always bleeding color. Even with less weight of the pictures.

  • Reconstruct only the block, mask it, invert mask, reconstruct and texturize : block and bleeding always there …

So I think we are reduced to mask pictures by hand based on mask from RC as layer … it seams that the masking option don’t work as it should (and if we try to realign after masking at the place of having a perfect component, even with the use of coded target and CP, we have mini components and 3 large components … that’s not logical).

I can share the dataset and printscreen of the problem but for us there is two bugs : near pixel blending from smthg that is removed from the mesh for texturing and big problems with masquerading (and there is no option of the mask use for texture).

Other thing, selecting all pictures, asking to delete the mask layer, saying in the console that all masks are deleted, going to the directory … all mask are still there … 

Dear ULB_Panorama,

it will be great, if you can share your data with us (trhond@capturingreality.com)

_deactivate the pictures that project this bleeding - _if you tried this, did you deactivated images completely or only for texturing?

Did you use masks from RC or from other application?

The mask were deleted from the project, but they still remain in the folder to later usage.

Hi,

we have deactivate image totally and in texture : but at this moment we lack of definition in this zone.

We have used masks, but as the support intersect the view we nneded to correct the statue masks (we have also made the support mask and invert it).

We found the problem with mask : if we edit them into Photoshop, photoshop seams to degrade the png and RC analyse it not well (instead of having a perfect contour, picture shown in RC is into a square …). So we scripted with opencv to open jpg source, transform it into png without compression and add the mask generated by RC into the alpha channel and replace the file extention into the rc project file in text mode (our script contain an option to invert or not the mask and with a dissolve option to extend the mask by a variable in px to avoid bleeding) : it’s working now perfectly but it was a pain in the ■■■ to understand the problem … png interpretation and possibility to rectifiy masks. 

Could be nice to add options about mask and texturing/colouring and perhaps generate mask based on depthmap to avoid problems with smtgh intersecting the view on some shot and avoid colour reprojection of thos unwanted parts.

Sincerly 

 

Hi,

after using mask, only masked parts of the image should be used for texturing and reconstruction.

For better understanding of your problem, can you please send us your data? Images, masks, project with errors…

I tested this mentioned problem on our data, but didn’t notice described problems…