search

jchibane / if-net

Implicit Feature Network (IF-Net) - Codebase
305 stars 59 forks source link

loss value #38

Open yjcaimeow opened 3 years ago

yjcaimeow commented 3 years ago

Dear author,

I retrained the shapenet model with spare point cloud as input. The initial loss values are about 20000. Is it normal since I think the loss is really large.

Do you remember the final loss value?

Best regards, Yingjie CAI

Bingoang commented 3 years ago

@yjcaimeow Hi, I'm also concerned about this issue. And what are your loss values in the later stage epochs?

XuHQ1997 commented 2 years ago

@Bingoang @jchibane I'm training Voxel Super-Resolution task based on this code, and also concered about this issue. It seems the loss values summed over all points of a shape, so the value is strangely large.

Have you trained this network? And what are your loss values after a long time training?

Looking forward for your reply. Thanks.

Bingoang commented 2 years ago

@XuHQ1997 Hi, in my training(within 100 epochs): For Voxel32, the minimum validation loss is about 12000; For Voxel128, the minimum validation loss is about 3200. The loss value looks very large, but the reconstruction looks reasonable.

YangZiz commented 1 year ago

Hello, I recently reproduced this paper. When I was training on my own training set, I ran a few epochs and had an overfitting problem. Have you ever had this situation？