On a practical side, normal maps are also fake detailing like bump maps, but they have 2 major advantages. It will display the effects of the normal map inverted on the mirrored side, and there is nothing you can do about it. So far, so good, but tangent space has issues when dealing with mirrored geometries, like the other side of a shoe/boot, the other hand of gloves, or the other side of the body. Now tangent space is used on organic models that bend and deform, like bone-rigged characters. Object space is used on static models that don't bend or deform, like 3D props. Which one to use depends on what you want the model to perform. Older rendering engines such as Firefly and 3Dlight use old school tangent space, and Poser also supports object space. There are basically 3 types of normal map: object space, tangent space, and MikkT space. These are the newest technology in the bunch, and again, works in a completely different way. In general, displacement maps require considerable computer power to process, so they impact rendering performance considerably. If it doesn't, the geometry will require a lot of additional faces to be able to produce the level of detail required by the displacement map. If it does, even a single face can be greatly detailed with a displacement map, even if it doesn't have the sufficient geometry to do it. What kind of geometry can be displaced depends if the rendering engine supports (or not) microface displacement. They actually displace the geometry (hence the name) to create physical bumps instead of faking their shadows. The math to calculate the normal face vector direction has some level of impact on rendering times.Дisplacement maps work in a completely different way. It can fool the eyes in one way, but not the other. So even if the model may appear to have bumps, its silhouette will remain perfectly smooth. So the effect depends on lighting directions, and cannot be seen from outlines of the 3D model. It instead fakes it by means of mathematically changing the face's normal vectors to cause fake shadows that simulate the presence of bumps. In addition to these limitations, bump maps do not really change the geometry shape. In other words, in the direction of the surface normal vector. But wait, there is more - bump maps can only represent bumps in a single direction, perpendicular to the surface. And by 8-bits, I mean the greyscale maps can only use up to 256 levels of grey, which limits the smoothness of the results. Bump maps are 8-bit greyscale maps that add bumpiness to 3D surfaces. Let's cover all 3 kinds of maps right here! But before I cover each in detail, let's remember that they all serve the same purpose, but in different ways. People ask me about this so often that I have decided to make the whole nine yards clear in one place. So the following answer is not for real displacement, it's only to fake 3D with Normal or Bump maps, in that case it's much less important if the grey spectrum doesn't respect exactly the heights of the real object.There seems to be a considerable level of confusion about what's the difference between bump, displacement and normal maps, and even some dispute about which is better. I chose a coin that had a better chance to give a good result: no oblique shadows, and the dark areas can be interpreted as the the bottom surface, bright areas as the peaks. Here is what my attempt gave from you coin picture. I don't know if you can achieve an accurate real displacement just from a b&w photo, because for this purpose you will need a picture that will precisely use black for bottom surfaces, white for peak surfaces, and all the shades of grey inbetween respecting the heights of the object surfaces, which is rarely the case with most photos. Actually it is still not clear to me if you want to fake the 3D effect or if you want to really displace the topology, for example for 3D print purpose.

Displacement map vs normal map