Texture "pass through" to occluded faces when texture mapping

[Woolseyyy]

Describe the bug It seems that the texture color pass through the front faces and map to both the front faces and the occluded faces when texture mapping. For example:

[cdcseacave]

The occlusion is normally taken into account; pls share debug data (scene and images)

[Woolseyyy]

The occlusion is normally taken into account; pls share debug data (scene and images)

https://www.dropbox.com/s/bsd8jgm17r01dys/hotdog_my.7z?dl=0

[Woolseyyy]

I have another problem... My purpose is to map recovered diffuse albedo and roughness map to a given mesh. However, a new obj would be produced at TextureMesh. What' s worse, I find that if I run TextureMesh with the same command and different images twice, the uv of the produced mesh is different. Hence, my roughness map and diffuse albedo cannot be shared with the same mesh...

[Woolseyyy]

The occlusion is normally taken into account; pls share debug data (scene and images)

I think I find the reason. It seems there are tinny bias of my mesh or camera poses so when I map texture, a little texture near the boundary of the mesh would pass through. At the same time, I set --global-seam-leveling and --local-seam-leveling as 1 so the middle area would be blended. Hence, a big area is "passed through".

[cdcseacave]

nor sure what you mean by pass through, I did run texturing on my machine and all looks good; how do you know it is wrong?

pls use latest develop branch, and the command line u use should simply be:

InterfaceCOLMAP .
TextureMesh scene.mvs --mesh-file scene_mesh.ply

where scene_mesh.ply is the known mesh

As for different uv mapping, I think that is caused by multi thrreading, pls try using only one thread --max-threads 1

[Woolseyyy]

nor sure what you mean by pass through, I did run texturing on my machine and all looks good; how do you know it is wrong?

pls use latest develop branch, and the command line u use should simply be:

InterfaceCOLMAP .
TextureMesh scene.mvs --mesh-file scene_mesh.ply

where scene_mesh.ply is the known mesh

As for different uv mapping, I think that is caused by multi thrreading, pls try using only one thread --max-threads 1

Can you have a check with "roughness" folder? I meet the problem mainly with the roughness data. Thank you about the answer about uv mappling! I will have a try.

[cdcseacave]

the algorithm is the same regardless of what the images contain

[Woolseyyy]

--max-threads 1

ok... In my results, it still remains some "pass through" at the bottom of bread:

[Woolseyyy]

nor sure what you mean by pass through, I did run texturing on my machine and all looks good; how do you know it is wrong?

pls use latest develop branch, and the command line u use should simply be:

InterfaceCOLMAP .
TextureMesh scene.mvs --mesh-file scene_mesh.ply

where scene_mesh.ply is the known mesh

As for different uv mapping, I think that is caused by multi thrreading, pls try using only one thread --max-threads 1

I tried --max-threads 1 in InterfaceCOLMAP , ReconstructMesh , and TextureMesh, but the uv issue still remains.

[cdcseacave]

again, how do u know is not seen by an image, for ex this one

[cdcseacave]

u do not need ReconstructMesh

[Woolseyyy]

InterfaceCOLMAP

I use

InterfaceCOLMAP .  --max-threads 1
TextureMesh scene.mvs --mesh-file scene_mesh.ply  --max-threads 1

with the latest realse version https://github.com/cdcseacave/openMVS/releases/tag/v2.0.1 but still the uv issue remains.

[Woolseyyy]

InterfaceCOLMAP

I use

InterfaceCOLMAP .  --max-threads 1
TextureMesh scene.mvs --mesh-file scene_mesh.ply  --max-threads 1

with the latest realse version https://github.com/cdcseacave/openMVS/releases/tag/v2.0.1 but still the uv issue remains.

I will have a try with https://github.com/cdcseacave/openMVS/pull/892

[Woolseyyy]

u do not need ReconstructMesh

I try with the latest code in 'develop' branch, and use

InterfaceCOLMAP .  --max-threads 1
TextureMesh scene.mvs --mesh-file scene_mesh.ply  --max-threads 1

but the uv issue still remains... The data and results are https://www.dropbox.com/s/zf0xhniveqbdsu3/hotdog_my.7z?dl=0

[Woolseyyy]

About the "pass through", I think there still exsit some problems. Have a look at this case:

I think the clues are clear in this input image:

However, the recovered results look like that the apple is projected to the plane.

The data and results are in https://www.dropbox.com/s/6hgapbjtfddmfjy/fruit.7z?dl=0

[cdcseacave]

I agree, something is not right here; do you have time to look into what images are projecting there bellow the objects? you have to check ListCameraFaces() where a list of images seeing each triangle is made; check some of the triangles that have wrong texture by which images are seen

[cdcseacave]

I found the problem in this case: the mesh has some triangles too big; I remeshed it and the result is ok scene_mesh.zip

[Woolseyyy]

I found the problem in this case: the mesh has some triangles too big; I remeshed it and the result is ok scene_mesh.zip

I see. Thank you very much!