How to project (or paste)panorama to model?

After I ask this question ,somebody tell me that what I want maybe is cubetexture(texturecube)? I am confuse that cubetexture is not designed for cubebox(hexahedron) only??!
Following are Initial question:
Recently,I came upon a unique way to get fluent 3D roaming experience on matterport’s Official network

I just want to know how did they do that because theire product is very fluent when swich the panorama picture.

After I roaming many times,I found that the panorama carrier they use is not box or sphere,but is the low quality model they show first!The evidence is that when switch the point,the object such as chair and table would have their own shadow(one chair have two image one stand up and the other one lie on the floorenter).With the object in panorama paste on their own correspond object and with depth information the roaming switch become more fluent (As for why they do not use the object directly ,I think because of the limited hardware,Many irregularity faces which get from scanning equipment cannot be use directlyenter )

And I want to use in my demo ,I have a group of six panorama which can paste on a boxGeometry perfectly,and I just want to paste them on model(low quality).but I stuck in project 360 degree.Yes I just find how to project one direction but I cannot project the remaining five.

var _p=BufferGeometry.attributes;
for (var i = 0; i < _p.position.count; i++){
   var uvtempbeforeconvert= ( new    THREE.Vector3(_p.position.array[3*i],_p.position.array[3*i+1],_p.position.array[3*i+2]) ).clone().applyMatrix4(houseObject.matrixWorld).project(camera1)
//use the worldvertices to get its screen coordinate            
  if(uvtempbeforeconvert.x<1&&uvtempbeforeconvert.x>-1 && uvtempbeforeconvert.y<1 &&uvtempbeforeconvert.y>-1)      {
//change the uvof vertices in right frustrum

Currently I can only calculate one direction.BUT I can‘t deal with the triangle faces which occupy two or more view frustum,like a face at the edge of the box.

How should I deal with this problem?Or I run in the FALSE direction at first?Which direction should i run in ?
Thanks in advance!

They have developed a 3D depth camera specially for capturing panos in one process, the scanned environment (point clouds) are triangulated and textured with the images. The panoramas are basic cube textures, the 3D part is only to get smooth transitions and the feel of naturalling moving between spots instead teleporting. It is very proprietary and quite expensive though, with a company specialized on panos we decided to not use it for some more reasons than the price and restrictive guidlines.

I remember in their demos you can switch into the 3D scan only, you will see that everything the camera can’t see will be non-existent, but it’s quite enough for a better experiene in transitions. This approach requires a special depth camera for capturing panos and depth at the same spot. In google panos it’s rather varying i guess, since they partially had depth information but also just somehow “wraped” between spots to make it look more fluent. Matter port was bought by google btw. (what else…)

After writing all this i’m not quite sure if you’re asking about their approach or how to make a basic pano. For that there are different mapping methods like sphere, cylinder or cube, while cube is rather recommended with least distortion in all directions and only 12 faces. There is a basic cube panorama in the examples.

1 Like

Thank you for your wonderful answer.Your penetrating analysis make me know more about their skills.And I just want to know how should I do to acheive similar effect when I already have six panorama pictures and a low quality(without using equipment but with Modeling software):sweat_smile: