search

XuyangBai / FoldingNet

PyTorch reimplementation of FoldingNet
35 stars 8 forks source link

ChamferLoss seems wrong to use? #1

Closed machengcheng2016 closed 4 years ago

machengcheng2016 commented 4 years ago

class ChamferLoss(nn.Module): def forward(self, preds, gts): P = self.batch_pairwise_dist(gts, preds) But when using ChamferLoss, the scripts are: output = self.model(pts) loss = self.model.get_loss(pts, output) Which seems just opposite

XuyangBai commented 4 years ago

Hi @machengcheng2016 Yes, you are right. This should be a mistake.

machengcheng2016 commented 4 years ago

Two negatives make a positive, I guess? The codes still run correctly...

XuyangBai commented 4 years ago

Yes, because for the Chamfer loss, the roles of two point clouds can be exchanged. So it still works well.