search

hansen7 / OcCo

[ICCV '21] "Unsupervised Point Cloud Pre-training via Occlusion Completion"
https://hansen7.github.io/OcCo/
MIT License
228 stars 26 forks source link

Pre-trained weights for DGCNN #5

Closed NCTU-VRDL closed 3 years ago

NCTU-VRDL commented 3 years ago

Hi!

Would you share the weights of DGCNN model? Thank so much!

I am also wondering that why can't I pre-train the model on scanobjectnn dataset?

hansen7 commented 3 years ago

Hi @NCTU-VRDL, thanks for the interest in our work!

We will release the pre-trained model as well as formalising the test scripts and part segmentation very soon.

One of the reason not to use ScanObjectNN for completion training is because of lack of full object shape therefore it is a bit skeptical to calculate Chamfer distance loss. But it might be useful for downstream tasks as well. You can also take a look at the failed completion objects in readme :)

NCTU-VRDL commented 3 years ago

Hi!

Thanks for your reply. I am currently working on a point cloud segmentation project and wish to submit our work to CVPR2021. I hope OcCo pre-training methodology can be included in our experiments. Look forward to the pre-trained model. Thanks so much!

NCTU-VRDL commented 3 years ago

Hi @hansen7!

Sorry for bothering you. Is there any updates for the pre-training weights? I only need that for visualization. Thank you:)

hansen7 commented 3 years ago

Hi @NCTU-VRDL, sorry for the lagging response, will upload by the end of this weekend ;)

hansen7 commented 3 years ago

Hi @NCTU-VRDL, please see it here, hope it catches ;)