some questions about the implement of CSPN
yudmoe opened this issue · 4 comments
Hi， thank you for you brilliant code, it's very helpful.
I have a question about the sample and stitch steps of model PENet_C2_train in the model, sample operation in line 921 and stitch operation 965.
Where can I read the paper of it, I noticed that #16 ,#59 some issue called it DA-CSPN++, but I did not find the paper. Is it a variant of RA-CSPN in cspn++?
the second question is about the propagation step (forward) of function CSPN and CSPNAccelerate, I notice that not only did you input the guide and depth (hn), but you also included the original depth (h0) as input, and the center guide is used in the the original depth (h0), which is consistent with the original CSPN paper (the Eq.1 of 3.1).
However，the CSPN code did not do so ( or maybe i didn't find it).
I'm just curious how this two implements are done affects the outcome, and what experience do you have with this
No，I got some answer of my second question in my own experiment， take original depth (h0) into the CSPN will slightly improve the accuracy
…
------------------ 原始邮件 ------------------ 发件人: @.>; 发送时间: 2023年2月13日(星期一) 中午1:25 收件人: @.>; 抄送: @.>; @.>; 主题: Re: [JUGGHM/PENet_ICRA2021] some questions about the implement of CSPN (Issue #60) Hello, have you solved your problem? I have the same question. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>
我们能加一个联系方式吗？