Added Matlab script to convert the BFM to a raw binary file

share/convert_bfm2009_to_raw_binary.m

+% Converts the 2009 Basel Face Model (BFM, [1]) to a binary file that can be
+% directly read byte for byte in C++. The workflow is to feed this
+% generated "raw" binary file into the bfm-binary-to-cereal app, which
+% reads this binary byte for byte and converts it to a cereal .bin file
+% that is then readable by eos.
+%
+% [1]: A 3D Face Model for Pose and Illumination Invariant Face
+% Recognition, P. Paysan, R. Knothe, B. Amberg, S. Romdhani, and T. Vetter,
+% AVSS 2009.
+% http://faces.cs.unibas.ch/bfm/main.php?nav=1-0&id=basel_face_model
+%
+% Developer notes:
+%  - The BFM data type is single, SFM is double
+%  - The BFM Matlab file contains the "unnormalised", orthonormal basis
+%    (as do the Surrey .scm files).
+%  - Domains:
+%    Colour: BFM: [0, 255], SFM: [0, 1].
+%    Shape: BFM: in mm (e.g. 50000), SFM: in cm, e.g. 50.
+%    (Note: I think that's wrong, since we have to divide by 1000.)
+%  - The BFM doesn't have any texture coordinates.
+%
+function [] = convert_bfm2009_to_raw_binary(bfm_file, binary_out_file)
+
+if (~exist('bfm_file', 'var'))
+    bfm_file = 'D:/Github/data/bfm/PublicMM1/01_MorphableModel.mat';
+end
+if (~exist('binary_out_file', 'var'))
+    binary_out_file = 'bfm.raw';
+end
+
+bfm = load(bfm_file);
+
+f = fopen(binary_out_file, 'w');
+
+fwrite(f, size(bfm.shapeMU, 1), 'int32'); % num vertices times 3
+fwrite(f, size(bfm.shapePC, 2), 'int32'); % number of basis vectors
+
+% Write the shape mean:
+for i=1:size(bfm.shapeMU, 1)
+    fwrite(f, bfm.shapeMU(i), 'float');
+end
+
+% Write the unnormalised shape PCA basis matrix:
+% All of basis 1 will be written first, then basis 2, etc.
+for basis=1:size(bfm.shapePC, 2)
+    for j=1:size(bfm.shapePC, 1) % all data points of the basis
+        fwrite(f, bfm.shapePC(j, basis), 'float');
+    end
+end
+
+% Write the shape eigenvalues:
+for i=1:size(bfm.shapeEV, 1)
+    fwrite(f, bfm.shapeEV(i), 'float');
+end
+
+% Write num_triangles and the triangle list:
+fwrite(f, size(bfm.tl, 1), 'int32');
+for i=1:size(bfm.tl, 1)
+    fwrite(f, bfm.tl(i, 1), 'int32');
+    fwrite(f, bfm.tl(i, 2), 'int32');
+    fwrite(f, bfm.tl(i, 3), 'int32');
+end
+
+% Now just exactly the same for the colour (albedo) model:
+fwrite(f, size(bfm.texMU, 1), 'int32'); % num vertices times 3
+fwrite(f, size(bfm.texPC, 2), 'int32'); % number of basis vectors
+
+% Write the colour mean:
+for i=1:size(bfm.texMU, 1)
+    fwrite(f, bfm.texMU(i), 'float');
+end
+
+% Write the unnormalised colour PCA basis matrix:
+% All of basis 1 will be written first, then basis 2, etc.
+for basis=1:size(bfm.texPC, 2)
+    for j=1:size(bfm.texPC, 1) % all data points of the basis
+        fwrite(f, bfm.texPC(j, basis), 'float');
+    end
+end
+
+% Write the colour eigenvalues:
+for i=1:size(bfm.texEV, 1)
+    fwrite(f, bfm.texEV(i), 'float');
+end
+
+fclose(f);
+
+end