Use SparseConvNet with C++ / tensorRT

[sjdsm]

Recently I'm working on a project trying to use a SparseConvNet model in a robotic system.
On one hand I tried to convert this python model to a TorchScript file. This seems to be a little bit difficult since SparseConvNet envolves a lot of special operations and control flows.
On the other hand I want to use tensorRT to speed up this model for real time performance. For this purpose I need to export this pytorch model to ONNX.
Can somebody kindly give me some suggestions?
SparseConvNet is much faster than common 3dconv but I got stuck in model exportation.

[daodaoling]

hello ! Have you solve the problem ?I meet the same problem .

[ltphy]

I think this one has not been supported yet. Issue 156

[daodaoling]

Yes,you are right !So I forgive it .

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人: "facebookresearch/SparseConvNet" <[email protected]&gt;; 发送时间:&nbsp;2020年8月13日(星期四) 上午10:22 收件人:&nbsp;"facebookresearch/SparseConvNet"<[email protected]&gt;; 抄送:&nbsp;"叨叨令"<[email protected]&gt;;"Comment"<[email protected]&gt;; 主题:&nbsp;Re: [facebookresearch/SparseConvNet] Use SparseConvNet with C++ / tensorRT (#170) I think this one havent been supported yet. Issue 156 — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or unsubscribe.