The way it works is you design the room points. You could (but i wouldnt suggest it) get a piece of graphed paper and design a 3d box which would be 1 x 1 x 1 squares in size, then you have to decide for yourself what the distance of 1 is. If you say that 1 square is = 10 meters then you have a box 10m x 10m x 10m, but that would have to remain the same for ALL objects in the game. Afterwards if you wanted to make a box 30m x 20m x 40m, you would have to stretch the 1 x 1 x 1 model that you made along each of its axises. Then you would get 3 x 2 x 4, and since you already decided that for your game 1 sqaure equals 10 meters, the box would be equal to 10(3) by 10(2) by 10(4).
Just remember that whatever you choose has to be the same and it doesnt really matter whether you make it a meter or a kilometer, because the height of your camera from, the floor will always be proportional to the world, in other words if the camera is 2 meters from the ground (roughly the height of a man) then the rest of the level can be made by real life measurements as well, but if you say that the camera will be 10 meters from the ground (5 times higher than normal) the rest of the world will also have to be made 5 times larger.
Once you have the points into your engine OpenGL or DirectX will be able to calculate the actual pixels depending on how you set up your camera.
Hope any of this helped, and sorry if I may have got of track a bit.