Bo Yang

Hi @usman151mrd, if you need to use pointnet++ as the backbone, it requires GPU implementation. You may use the simple pointnet as the backbone which is trainable on a cpu.

hi @zhaoyu19920930 , thanks for your interest in our paper. It's supposed to be able to run on tf 1.x versions, though you may modify slightly when necessary. The project...

Hi @zeng-ziyin please uncomment the lines here for visualization. https://github.com/Yang7879/3D-BoNet/blob/201ae46ffc6a00d5a9fae44da486d3702a83a1a3/main_eval.py#L212

Hi @maptastic @ZJU-PLP , it's very easy to save the results of each scene, just uncomment the following lines to check the results are correct and then save the variables...

Hi @maptastic , there might be some algorithms to merge the predicted bounding boxes according to the overlap, but I guess the bounding boxes per scene can be obtained once...

Hi @fxy2012 and @zhaoyu19920930 , the '.mat' file includes all the predicted semantic and instance results together with ground truth for each scene (room). Refer this line below to check...

Hi @pomelo93, 根据描述，你的数据集 和scannet 标准方式应该类似，对那些不感兴趣的点都直接当成背景，在training过程中，只是用你关注的objects去supervise。具体可以参考scannet dataset。

Hi @dsc1126, since we were following the online hidden testing for comparison, we actually didn't systematically evaluate the model on the val set and didn't have the figures at hand.

Hi @bonbonjour , be careful about the labels of scannet dataset, there are many pts not annotated.

Hi @lelouedec, the released code only deals with the case where the number of pred bbox is equal to the number of gt bbox. It's easy to slightly modify it...