0

I am learning 2D animations. I would like to know what is the easiest way to convert 2D images (given below) in to a single 3D image?

For example:

2D Images

3D Image

4 Answers 4

2

There is no easy way, 3 images is not enough to determine the shape of an object in 3D, unless its simply a very blocky object. Having more orthogonal projections does not guarantee anything. This then means you need to do some work, and human intuition, to decide how the object looks in the 3D projection.

If we take your first example image. We can project it in 3 directions, but it will not be very realistic as it has no roof and the shape of bonnet is not known. If fact there is no guarantee that the 3 pictures even work out in the projection

enter image description here

Image 1: When you start projecting the images you notice that it does not work out very well.

2
  • Not 3 Image, The side view of the car can be used for bothe side. So we have 4 sides can't we make them a single 3D model ? Commented Dec 28, 2016 at 8:01
  • @DasaradhMS No, it does not work out very well. The object must actually maker sense in 3D. When you project the image it stops making sense. Added quick image. So if you want the shape to make sense in 3d you porettymuch have to think it out in 3d the first time around. Too much detail thinking is missing form orthographic projections. If it were that easy we would do it all the time in mechanical engineering, but we dont because it does not work out very well.
    – joojaa
    Commented Dec 28, 2016 at 8:19
1

Photogrammetry theory covers how to convert 2D images to a 3D model. That theory is hardcore mathematics. Photogrammetry software is THE tool for making fantasy movies. Affordable PC applications are also developed such as Photoscan. It takes a bunch of photos. Those photos must cover whole object, at least 2 different photos of every surface point is needed. That means often 20 or even more photos taken from different directions. The object must not move and the light must be absolutely stable betveen the takes. Pro quality camera & lens is a must. No glossy surfaces are allowed because they do not present the object, but the surroundings. Live moving objects reguire say 20 simultaneous cameras and pro flashes for usable results. Photoscan and and the others solve the most probable surface that is the initial source for the photos. Generally the resulted 3D surface still needs plenty of manual editing in a 3D modelling program.

0

Smoothie 3D This free online software is a game-changer in the creation of a 3D model based on a single picture. With only one picture it helps you to create a simple 3D model online that looks closer to the kind of result that you get with a scan or photogrammetry.

2
  • Well yes but smoohie 3D is modelling application. With human input you can make nearly anything, even without a image. I mean i can do the same as smoothie 3D with nearly any modeling application even with, no image or with any number of images. It ha a nice gui tough.
    – joojaa
    Commented Dec 28, 2016 at 12:06
  • Smoothie's main trick is that nice qui. It really helps the beginner to step in, so low is the doorstep in that place.
    – user82991
    Commented Dec 30, 2016 at 22:24
0

I’ve done a rough model to exemplify some rules about lengths deformation in orthographic projections. Perspective is similar, only with extra deformation based on distance from the camera. The simplest way to project a plane is parallel to the projection plane. You get actual lengths of the object and therefor the actual ratio between verticals and horizontal segments. Note that the other 2 planes, like in a box, are perpendicular to the projection plane and have zero visibility. For example if you see the object (box) from the front , you get 100% visibility of that plane, but 0% of the top, bottom and sides. First step is to rotate the object in one axis, 45 degrees for convenience, like I did vertically. What you get then is a front and side that shrink in projection horizontally, verticals that remain actual lengths and a top and bottom view that have zero visibility. In my image I used a front-right orientation, but you can have a front-top, a side-top or any two neighbor planes. To see the object’s all three planes it must be rotated again horizontally, also 45 degrees, around the view axis, not the objects. That’s like a combination between a two plane projection, and the rotation in depth of a parallel plane, in this case the top view. The length from the highest points, if projected, will remain the same.

However, these are some guidelines that may only help the eye. In order to get precision it is necessary to use a trigonometric function: cosine = adjacent/ hypotenuse. You can find more details here:

https://math.stackexchange.com/questions/3284741/is-there-a-more-precise-way-to-determine-the-percentage-of-actual-length-of-a-pr

. The idea is that the actual length is the hypotenuse and the 45 degrees (or any given number) length is the adjacent. The cos function can be calculated using google calculator for example. Make sure you set it to degrees though, not radians. This works because in orthographic all projection lines are parallel to each other and perpendicular to the projection plane. (Note that at 45 degrees the projection is 70.7% of the actual length and it is 50% at 60 degrees). When all 3 planes of the object are visible, like in example, the verticals shrinks up to 70% at 45 degrees and so will the diagonal of a square compared to top view actual length diagonal. In the case of rectangle, like in the example, the diagonals get projected vertically first. enter image description here

So how about curved forms? My best guess is to find a way to associate it with boxes. First build a framework made of sections, then use that to figure out the contour. The sections will deform based on the angle, but the contour can stay consistent. For example a sphere will always project onto a circle, while an elongated sphere will merge between a circle and ellipses of different sizes.

Edit: I hope this other image about alignments of projections will help clarify the answer. If it doesn’t, just ignore it, next time I’ll keep it to myself. enter image description here

10
  • Err... How does this help. For one you dont usually calculate the trigonometrics in 2D drawing or 3D applications since that is what the transformation tools do for you
    – joojaa
    Commented Jan 11, 2020 at 6:23
  • @joojaa It’s helpful because it gives the right theory for once. All images are 2d images because you project form onto a plane, no matter what the view angle.
    – 3Dtoaster
    Commented Jan 11, 2020 at 6:45
  • No it does not. It tells .. really badly how isometric orthographic projection is formed. It does not tell at all how the shape is made. It does not tell how to do rotation etc. It also somehow assumes user has a 3D application.
    – joojaa
    Commented Jan 11, 2020 at 6:55
  • Also your answer dies not deal with, like at all with the implied assumption that the conversion can be made satisfactorily. No such guarantee exists.
    – joojaa
    Commented Jan 11, 2020 at 7:04
  • Anyway if you would stop out of theory and explain how you made the box version in your 3d App. That would be an answer.
    – joojaa
    Commented Jan 11, 2020 at 7:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.