search

XuyangBai / D3Feat.pytorch

[PyTorch] Official Implementation of CVPR'20 oral paper - D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features https://arxiv.org/abs/2003.03164
138 stars 20 forks source link

Demo code for D3Feat #13

Closed fabiopoiesi closed 2 years ago

fabiopoiesi commented 3 years ago

Hi,

Do you guys have a simple demo code, that given two point clouds, D3Feat computes the features and then these features are used to register the two point clouds with RANSAC?

Thanks

qsisi commented 3 years ago

I think for methods based on KPConv, it is not easy to directly apply it to single point cloud. As far as I know, KPConv needs to calibrate the appropriate neighborhood_limits first, so it's better to rewrite the Dataset and organize it to keep in same with that in D3Feat.

So all we need to do it's to rewrite the Dataset class for specific pointclouds and change the parameter first_subsampling_dl to be similar to the voxel_size of your own pointcloud data.

fabiopoiesi commented 3 years ago

I was actually thinking something simpler than that. Something like the do in FCGF. Given a pcd as input, get the features as output.

XuyangBai commented 2 years ago

Hi, sorry for the late reply. I did provide a demo script for the TensorFlow implementation. But as @qsisi said, since I use KPConv as the backbone, I need to pre-compute neighborhood_limits, the neighborhood indices, etc in collaten_fn, which is quite cumbersome so I still haven't written demo code for PyTorch. I can find time to write the code snippet for pytorch if you really need it.

fabiopoiesi commented 2 years ago

Hi, I actually managed to use the tensorflow implementation successfully. Thanks for your help.

On Wed, 5 Jan 2022 at 07:33, XuyangBai @.***> wrote:

Hi, sorry for the late reply. I did provide a demo script for the TensorFlow implementation https://github.com/XuyangBai/D3Feat/blob/master/demo_registration.py. But as @qsisi https://github.com/qsisi said, since I use KPConv as the backbone, I need to pre-compute neighborhood_limits, the neighborhood indices, etc in collaten_fn, which is quite cumbersome so I still haven't written demo code for PyTorch. I can find time to write the code snippet for pytorch if you really need it.

— Reply to this email directly, view it on GitHub https://github.com/XuyangBai/D3Feat.pytorch/issues/13#issuecomment-1005418989, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEBXWKDLEPPVHHDTSWPP4F3UUPQ2JANCNFSM5FNWY55Q . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you authored the thread.Message ID: @.***>

XuyangBai commented 2 years ago

Glad to here that:)

houyongkuo commented 1 year ago

Hi,

Do you guys have a simple demo code, that given two point clouds, D3Feat computes the features and then these features are used to register the two point clouds with RANSAC?

Thanks

Hi, I have finished a demo about 3dmatch dataset for PyTorch. demo

zouyajing commented 5 months ago

Hi, Do you guys have a simple demo code, that given two point clouds, D3Feat computes the features and then these features are used to register the two point clouds with RANSAC? Thanks

Hi, I have finished a demo about 3dmatch dataset for PyTorch. demo

thx a lot. do you have the D3Feat pytorch model backup? seem the google drive expire and the author is not taking care any more.