Table 2 Key hyperparameters (default unless stated)
From: HemaContour: explicit parametric contour learning for robust ICH segmentation on non-contrast CT
Setting | Value |
|---|---|
Backbone | U - Net (light), ConvNeXt-UNet (ablation) |
Spline degree/control points | d = 3 (cubic)/K = 64 |
Snake iterations/step size | T = 5/η = 0.1 |
Loss weights (λ1, λ2, λ3, λ4) | (1.0, 0.5, 0.1, 0.1) |
Optimizer/schedule | |
Base lr/weight decay | 3 × 10−4/5 × 10−4 |
Epochs/batch (slices/GPU) | 200/32 (accum. to 128 global) |
Mixed precision | AMP35 |
TTA | flip (2 views); EMA enabled |
