cmtk.streamxform and xform.default inconsistent with affine matrices

[jefferis]

scalemat=rgl::scaleMatrix(8/1e3,8/1e3,8/1e3)

pts=structure(c(23244, 23376, 23384, 22364, 23354, 4566, 23684, 23195.5,
              23728, 23750, 31140, 31690, 31306, 33230, 32284, 20140, 32966,
              33498.6, 31998, 32328, 26474, 26540, 25610, 26584, 25506, 8346,
              26034, 26241.1, 25616, 26122), .Dim = c(10L, 3L), .Dimnames = list(
                NULL, c("X", "Y", "Z")))

transforming points using the same affine matrix via CMTK and xformpoints.default results in different outcomes. One is doing the inverse of the other.

> xform(pts, reglist(scalemat, swap = T))
            X        Y        Z
 [1,] 185.952 249.1200 211.7920
 [2,] 187.008 253.5200 212.3200
 [3,] 187.072 250.4480 204.8800
 [4,] 178.912 265.8400 212.6720
 [5,] 186.832 258.2720 204.0480
 [6,]  36.528 161.1200  66.7680
 [7,] 189.472 263.7280 208.2720
 [8,] 185.564 267.9888 209.9288
 [9,] 189.824 255.9840 204.9280
[10,] 190.000 258.6240 208.9760

> xform(pts, reglist(scalemat, swap = F))
            X       Y       Z
 [1,] 2905500 3892500 3309250
 [2,] 2922000 3961250 3317500
 [3,] 2923000 3913250 3201250
 [4,] 2795500 4153750 3323000
 [5,] 2919250 4035500 3188250
 [6,]  570750 2517500 1043250
 [7,] 2960500 4120750 3254250
 [8,] 2899438 4187325 3280138
 [9,] 2966000 3999750 3202000
[10,] 2968750 4041000 3265250

> xform(pts, simplify_reglist(reglist(scalemat, swap = F), as.cmtk = T))
            X        Y        Z
 [1,] 185.952 249.1200 211.7920
 [2,] 187.008 253.5200 212.3200
 [3,] 187.072 250.4480 204.8800
 [4,] 178.912 265.8400 212.6720
 [5,] 186.832 258.2720 204.0480
 [6,]  36.528 161.1200  66.7680
 [7,] 189.472 263.7280 208.2720
 [8,] 185.564 267.9888 209.9288
 [9,] 189.824 255.9840 204.9280
[10,] 190.000 258.6240 208.9760

[jefferis]

This is not good news at all, since there must be numerous cases in the wild where we may rely on this behaviour. This includes

• mirror.default in nat
• elmr transforms
• I noticed the problem in nat.jrcfibf
• others?

[jefferis]

it looks like a significant overhaul of registration classes may be in order, to fix this, handle #387 and allow better mixing across different registration formats including:

• h5reg
• antsreg
• tpsreg

[jefferis]

I wonder if I should add this to the some of the updates planned for nat v2 (e.g. adding a base class for objects with 3d coordinates)