You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi, still me.
I have question about the MSE loss calculation, in the paper you mantioned that the loss only calculate on the frontier pixels, but I check the
mask = labels["loss_masks"][:, 2:]
# Evaluate predictions only on mask = 1 regions
mask_sum = mask.sum(dim=3).sum(dim=2) + 1e-16 # (b, 1)
The mask is a whole floor map not the frontier pixel by visualization, so you calaucte on the entier floor map instead of frontier pixels, am I correct?
Besides, why there is no mask processing on area_pf loss?
area_pf_loss = self.area_loss_fn(area_preds, area_gts).mean()
loss = loss + area_pf_loss
losses["area_pf_loss"] = area_pf_loss.item()
The text was updated successfully, but these errors were encountered:
Hi, still me.
I have question about the MSE loss calculation, in the paper you mantioned that the loss only calculate on the frontier pixels, but I check the
The mask is a whole floor map not the frontier pixel by visualization, so you calaucte on the entier floor map instead of frontier pixels, am I correct?
Besides, why there is no mask processing on area_pf loss?
The text was updated successfully, but these errors were encountered: