Does anyone know how I one might go about decomposing the matrix emitted by anisotropic mag estimation into rotation, scale, skew? It’s been a long while since I’ve done lin alg, but my old notes on decomposing 2x2 transform matrices is not giving me results that make sense.
Thanks!
-Craig
Hi @yoshiokc,
I believe the singular value decomposition can be used to decompose the anisotropic magnification matrix into a rotation, scaling (by 2 independent factors along orthogonal axes), and a subsequent rotation.
Note that before doing SVD on the anisotropic magnification matrix, you’ll have to pull it from particle dataset’s ctf/anisomag entry, reshape it into a shape of (2,2) (C-order), and then add it to the 2x2 identity matrix — this is because CryoSPARC stores the matrix with the identity subtracted off, for convenience reasons.
Best,
Michael
oooohh, identity is subtracted? that makes more sense what I’m seeing… in the past I’ve used the QR decomposition to get these values… I was alos copying the values out of the log- did not check to confirm if the values in the log and the particles are the same