"""Own C3 body plus zero-initialized distant OCT inputs; no metadata features.""" import torch from torch import nn from torch.nn import functional as F from src.models.context import ContextBoundaryNet class WideBoundaryNet(ContextBoundaryNet): def __init__(self, n_boundaries, base, prior_height): super().__init__(n_boundaries, base, prior_height) self.outer_conv = nn.Conv2d(2, base, 3, padding=1, bias=False) nn.init.zeros_(self.outer_conv.weight) def forward(self, image): if image.ndim != 4 or image.shape[1] != 5: raise ValueError("Five OCT input views required") local = image[:, 1:4].contiguous() outer = image[:, [0, 4]].contiguous() first = self.down[0].net x = first[0](local) + self.outer_conv(outer) for layer in list(first)[1:]: x = layer(x) skips = [x] for layer in self.down[1:]: x = layer(F.avg_pool2d(x, 2)) skips.append(x) for index, layer in enumerate(self.up): skip = skips[-2 - index] x = F.interpolate( x, size=skip.shape[-2:], mode="bilinear", align_corners=False ) x = layer(torch.cat([x, skip], 1)) logits = self.head(x) prior = self.axial_prior if prior.shape[-2] != logits.shape[-2]: prior = F.interpolate( prior, size=(logits.shape[-2], 1), mode="bilinear", align_corners=False ) return logits + prior