WisconsinAIVision/MaskPoint

About the code release

Closed this issue · 5 comments

ruili3 commented

Hi,

I read the paper and the idea is quite impressive!

Since sharing the complete training & testing code takes time, can you share the architectures of the network for a first glance? We're conducting some related research and wish to learn from it.

Thanks!

haotian-liu commented

Sorry for the late reply -- I somehow missed this issue. We have just released the code and the checkpoints.

And for the architecture of the backbone, we mainly follow the Point-BERT design with plain Transformers for a fair comparison.

WWW2323 commented

@haotian-liu hi, thanks for your codes very much! And, when will the codes about ScanNetV2 come? thanks~

haotian-liu commented

@WWW2323 Thank you for your interest in our project. Our plan is to release the ScanNetV2 code and checkpoints by early next week.

haotian-liu commented

@WWW2323 The 3D object detection code has just been released :)

WWW2323 commented

Great! And, how long did you train 3detr? Did you use only one GPU to train 3detr for 1080 epochs? It seems to need a week for me, orz