The PhotoFit is a multithreaded CPU-bound process. It uses all available hardware cores and scales well with multiple cores. It does not use the GPU. It runs fastest when built 64-bit.
Required RAM is 40MB plus several times the combined size of the input photos.
Updating vertex positions for an SSM requires each vertex of the mean model to be perturbed by some factor of each of the 80 modes. This requires V*240 float multiplies and V*240 float additions.
This should be followed by surface normal generation. Pre-calculated surface normals are not recommended for SSMs as they tend to mask changes in shape.
Constructing a color map from an SCM requires each pixel of the mean texture to be perturbed by some factor of each of the 50 modes. This requires P*3*50 fixed-point multiplies, additions and fixed point to float conversions, as well as a LUT for gamma correction. Application of a detail texture requires resampling it into the UV layout of the map, and possibly upscaling the generated map to match the resolution.
An .EGM file requires 6 bytes for each mesh vertex for each of the 80 shape modes. Thus a head model of 5000 verts have an .EGM file of about 2.3MB. The memory footprint is twice that.
An .EGT file requires 3 bytes per pixel for each of the 50 modes. The head map should be 512x512 for a total size of about 40MB. Body color stats can be much smaller since the color is constant below the neck, so their storage is negligable.
Each face is defined by its face coordinate, which is 130 numbers, plus an optional JPG-encoded detail texture.