rounding eval (which I'm not sure I believe in as a concept)

full resolution, no finetune:

mTRE: 2.311035590485163, mTRE_X: 0.873484864263977, mTRE_Y: 1.1453686289447025, mTRE_Z: 1.398816546268779, DICE: 0.9839830646482749
mTRE: 1.7008697594492386, mTRE_X: 0.6436288430159421, mTRE_Y: 0.8176459010038764, mTRE_Z: 1.0994405529901972, DICE: 0.9875815784128983
0.9839830646482749

full resolution, finetune

mTRE: 1.478871996309246, mTRE_X: 0.5785209564705847, mTRE_Y: 0.7231673335360019, mTRE_Z: 0.9272361422043595, DICE: 0.987072093794362
mTRE: 1.1395389561975062, mTRE_X: 0.46966112596329185, mTRE_Y: 0.5532936684554987, mTRE_Z: 0.7847049879382644, DICE: 0.989562143510095
0.987072093794362

half res, no finetune:

mTRE: 2.709416346293769, mTRE_X: 0.9923500632764851, mTRE_Y: 1.3895394382073543, mTRE_Z: 1.5801930664408113, DICE: 0.9704969758656941
mTRE: 2.047477228986998, mTRE_X: 0.7513863519073622, mTRE_Y: 1.0437182204639481, mTRE_Z: 1.2435612909202693, DICE: 0.9784818002348021

half res, finetune:

mTRE: 1.898301318435239, mTRE_X: 0.7189809711008761, mTRE_Y: 0.9942497389745215, mTRE_Z: 1.1031087761070628, DICE: 0.9741556898715056
mTRE: 1.4538276066290883, mTRE_X: 0.5693061373044725, mTRE_Y: 0.7449523584196344, mTRE_Z: 0.909844917377109, DICE: 0.9802551099407191

Experiments for paper

We have

Lung with no augmentation, adaptive loss on NCC Lung with augmentation, LNCC

Knee with no augmentation, MSE, Knee with no augmentation, LNCC

Knee with no augmentation, ICON, MSE Downsam

Re-run all experiments with lung structure