search

mikacuy / pointnetvlad

PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition, CVPR 2018
MIT License
351 stars 72 forks source link

Is the loss trend normal? #15

Closed cloudcjf closed 3 years ago

cloudcjf commented 3 years ago

@mikacuy Hi，mika. First of all, thanks for your great work. I'm new to place recognition tasks and trying to retrain your network as a tutorial. I used to retrain some other networks and the losses go almost down directly while the loss defined in your network keeps going up and down. Here is a screenshot which I log out the batch losses. So does it go the right way as you did before? Tell me if I miss any details. Thanks image

Claude Cui

Master-cai commented 1 year ago

hi, I have the same question, do you have any idea about that? thanks!

cloudcjf commented 1 year ago

hi, I have the same question, do you have any idea about that? thanks!

The place recognition task adopted the triplet loss and Hinger loss to backpropagate the parameters. The zero-loss means that the distance between the anchor and positive point cloud in this tuple is already less than the anchor and negative point cloud.

Master-cai commented 1 year ago

hi, I have the same question, do you have any idea about that? thanks!

The place recognition task adopted the triplet loss and Hinger loss to backpropagate the parameters. The zero-loss means that the distance between the anchor and positive point cloud in this tuple is already less than the anchor and negative point cloud.

thanks a lot!

If you don't mind, I have another question: I find in the file "train_pointnetvlad", the global variable "HARD_NEGATIVES={}" is never updated which means it is useless, is there something wrong here?

cloudcjf commented 1 year ago

hi, I have the same question, do you have any idea about that? thanks!

The place recognition task adopted the triplet loss and Hinger loss to backpropagate the parameters. The zero-loss means that the distance between the anchor and positive point cloud in this tuple is already less than the anchor and negative point cloud.

thanks a lot!

If you don't mind, I have another question: I find in the file "train_pointnetvlad", the global variable "HARD_NEGATIVES={}" is never updated which means it is useless, is there something wrong here?

Yes, if you just follow the default settings, it will never be updated