I have a task, in which I have to take in 3 input images. For it I wrote a custom u-net model
def __init__(self, n_channels, n_classes, bilinear=True): super(ThreeInputs, self).__init__() self.n_channels = n_channels self.n_classes = n_classes self.bilinear = bilinear self.inc = DoubleConv(n_channels, 64) self.down1 = Down(64, 128) self.down2 = Down(128, 256) self.down3 = Down(256, 512) factor = 2 if bilinear else 1 self.down4 = Down(512, 1024 // factor) self.up1 = Up(1024, 512 // factor, bilinear) self.up2 = Up(512, 256 // factor, bilinear) self.up3 = Up(256, 128 // factor, bilinear) self.up4 = Up(128, 64, bilinear) self.outc = OutConv(64, n_classes) def forward(self, x1, x2, x3): x = torch.cat([x1, x2, x3]) x1 = self.inc(x) x2 = self.down1(x1) x3 = self.down2(x2) x4 = self.down3(x3) x5 = self.down4(x4) x = self.up1(x5, x4) x = self.up2(x, x3) x = self.up3(x, x2) x = self.up4(x, x1) logits = self.outc(x) return logits
I passed it to a learner:
learn = Learner(dls, ThreeInputs(3,2), opt_func=ranger, metrics=acc_camvid)
where dls is:
field = DataBlock(blocks=(ImageSequenceBlock, MaskBlock(codes)), get_items=SequenceGetItems([0,1,2], -1), splitter=RandomSplitter(), get_x=get_x, get_y=get_y) dls = field.dataloaders(fns_train, bs=2, create_batch=create_batch, verbose=True)
I am pretty sure that when I used to pass custom models to Learner before it would deal with summary and lr_find on it own. Wouldn’t require the custom model to have it. However, now it does.
My question is how to I overcome it?