Closed 111sadas closed 1 month ago
Hi @111sadas, thank you for interest in our paper! To support occupancy output, you can add modify self.fc_out to a linear layer and a sigmoid function which outputs a value between 0 - 1, then, use cross entropy loss (example) to train the occ prediction.
Great work! When I tried to train the Slice3D network using our dataset, I changed the parameter "pred_type" to "occ" and found that the network framework lacked the "occ_pred" output. How should I operate to implement the 'occ' method of training?