Question about SVM training.

[zeal-up]

Hi, thanks for your job. I have some question about the svm training and testing precedure. According to your paper, you randomly rotate the pointcloud and extract the features. Dose it mean that each you only extract only one features for each pointcloud?(for example, you have 1k pointclouds, you extract 1k feature vectors and send them to the svm?)

The next question is, do you apply any test augmentation during svm testing?

[optas]

Hi,

Yes, we extract 1 feature (the bottleneck of the AE) per point-cloud. We don't apply any test-augmentations. I will try to release the notebook/code soon. Thanks a lot for your interest.

[moromatt]

Hi,

Yes, we extract 1 feature (the bottleneck of the AE) per point-cloud.

Hi optas!

When you talk about "1 feature per point-cloud", you mean 1 feature vector per point-cloud? (e.g. 100 chairs point-clouds with 128 feature per point-cloud, so 100x128 features)

[optas]

On Sep 9, 2019, at 5:59 PM, Matteo Moro <[email protected]mailto:[email protected]> wrote:

Hi optas,

Hi,

Yes, we extract 1 feature (the bottleneck of the AE) per point-cloud.

Hi optas!

When you talk about "1 feature per point-cloud", you mean 1 feature vector per point-cloud? (e.g. 100 chairs point-clouds with 128 feature per point-cloud, so 100x128 features)

Yes.

— You are receiving this because you commented. Reply to this email directly, view it on GitHubhttps://github.com/optas/latent_3d_points/issues/16?email_source=notifications&email_token=AA4SRJCA4MJHW4MCZSTBSGDQI3WQ7A5CNFSM4GB7TWA2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6JPBRQ#issuecomment-529723590, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AA4SRJHTF63GNVUQJGGIWALQI3WQ7ANCNFSM4GB7TWAQ.

[zshyang]

Hi @optas ,

It is might be too late to ask since this project is about 4-5 years ago. But if it would be possible for your to release the part of code for SVM. I totally understand your decision.

Best regards.