huge loss

[domadaaaa]

when i run voxelmorph with my dataset using NCC loss，a while after，NCC loss up to thousand or bigger，how can i solved it？
output like this
i: 1 loss: -0.092982 sim: -0.092983 grad: 0.000000
i: 2 loss: -0.095556 sim: -0.095647 grad: 0.000023
i: 3 loss: -0.104686 sim: -0.105146 grad: 0.000115
i: 4 loss: -0.179067 sim: -0.180766 grad: 0.000425
i: 5 loss: -0.242327 sim: -0.245312 grad: 0.000746
i: 6 loss: -0.331464 sim: -0.333647 grad: 0.000546
i: 7 loss: -3.075087 sim: -3.082510 grad: 0.001856
i: 8 loss: -80.884193 sim: -80.894211 grad: 0.002504
i: 9 loss: -2233.514404 sim: -2233.535645 grad: 0.005287
i: 10 loss: -0.641073 sim: -0.645906 grad: 0.001208
i: 11 loss: -330.219147 sim: -330.225281 grad: 0.001531
i: 12 loss: -13.687098 sim: -13.695247 grad: 0.002037
i: 13 loss: -0.485161 sim: -0.491218 grad: 0.001514
i: 14 loss: -19.163136 sim: -19.171404 grad: 0.002067
i: 15 loss: -1.049533 sim: -1.056095 grad: 0.001640

[Pull2MAX]

你归一化了吗

[domadaaaa]

你归一化了吗

归一化了，我感觉是数据的问题，换了个数据就没问题了