The structure-from-shading problem is the problem of knowing the shape of a 3-D object from seeing how light falls on it; it suffers from a number of difficulties such as having to assume that the object is all the same color and not being able to tell when a Lambertian surface rotates around the vector of the [dominant] light beam.
But if the object is covered in or made of woven or nonwoven cloth, as the humans often are, and you have a high-enough-resolution image to see the individual threads of the cloth (or meltpoints, in the case of nonwovens like friselina), you have several great advantages.
First, the threads or meltpoints give you a dependable, repeating three-dimensional microtexture, which gives you a sample of illumination of different surface normals in the neighborhood of the overall surface normal.
Second, thread or meltpoint-line direction discontinuities often indicate surface discontinuities. (Sometimes they’re just seams, though.)
Third, the threads or lines of meltpoints are typically at right angles, and this gives you an independent clue about how inclined each surface patch is to the camera.
Fourth, the threads or meltpoints are typically evenly spaced over the entire surface, and this allows estimation of relative Z-coordinates by perspective.
Fifth, although fabric is somewhat flexible, the threads or meltpoints generally run near ecliptics over the surface if the fabric is to not wrinkle. Nonwoven fabric is much less flexible.
Some fabrics, such as plaids, stripeds, houndstooths, and certain prints, are amenable to algorithms based on some of these properties even if the image resolution is not good enough to see individual threads.