Ok so I am stuck trying to figure out how to feed my targets for a multi target regression.
The data is loaded using a dataloader from a custom data class:
data = DataBunch(train_dl=train_dataloader, valid_dl=val_dataloader, device=torch.device(device))
The targets are coordinates within a bounding box of the format (batchsize, 2, 1) where each value corresponds to a coordinate eg:
tensor([[0.5244511366],
[0.4735262394]])
When I calculate the loss I get one value for each prediction a mean of both values. This causes the network to learn the centre of my bounding box which makes sense but is not really useful.
So the question is how can I calculate separate losses for each X and Y coordinate?
Any help in getting myself unstuck would be appreciated!