Kuangen Zhang

Hi, thanks for your interest. Yes. we used the same EdgeConv layer as in DGCNN. In our paper, we just showed why the EdgeConv itself could extract multi-view features, which...

Hi, thanks for your question. K is a super parameter and we cannot know the best value for different datasets. It may depend on the density of the point cloud....

Hi: I have also solved bug of PclAlgo, please see my update in the commits 713a2688ca4b892357812e5e717ef1941359ddb9 or just pull my code through: git pull. Thanks for reporting this issue.

Hi, thanks for your interest. The reason is that linking hierarchical features may avoid vanishing gradient problem, which is a similar idea as ResNet.

Thanks for reporting this issue. I have just run the evaluate.py on my computer. However, I don't find this issue on my computer. I am not sure if it is...

Thanks for reporting this issue. I have updated the path in the code, please see my new commit: e58017a5436738c159376c44ffb91930685990a2

> acc can't reach at 93.%;this code i have run two times,i don't change any parameters. Thanks for your message. I will check the code recently. Moreover, how much did...

> @KuangenZhang Hi,I run the baseline,and I got the result of 92.2% >  Hi, I will run the code again and check this issue. However, it is possible that...

 @grxiao @Yuquan-Lu Hi, I have run the code again and shown the results in the above picture. The highest classification accuracy is 92.73%. The accuracy does vary around 92.5%....

@Yuquan-Lu @grxiao Moreover, I changed the batch_size to 16, which seems to increase the stability of the training result.