northsummer / lgkd Goto Github PK

Hi, thanks for the work. I wonder if voxelisation is done first followed by projection, or the other way around? The text description and the figure seem inconsistent on this. Also, I'd appreciate if you explain a bit more on their implementation. Thanks.

Hi, I read your paper on Arxiv and the performance is very impressive. I wonder if the ResNet-18 you used was pretrained on other datasets before (ImageNet or other detection datasets). Looking forward to your reply~