search

grimoire / torch2trt_dynamic

A pytorch to tensorrt convert with dynamic shape support
MIT License
254 stars 34 forks source link

is this library supposed to be compatible with original torch2trt library? #6

Closed gengshenghong closed 3 years ago

gengshenghong commented 3 years ago

is this library supposed to be compatible with original torch2trt library? As the library use same package name with original one, we can only install one of them. The issue is, when I use this dynamic version on Jetson Nano, the Nvidia trt_pose (who depends on torch2trt library) example (https://github.com/NVIDIA-AI-IOT/trt_pose) does not work well.

I am not very familiar with these libraries, please help to confirm if it supposed to be compatible before looking into details. If yes, I can paste the errors here. If no, it may be an issue to override the official torch2trt library and any suggestions?

Thanks, Shenghong

grimoire commented 3 years ago

Hi Thanks for the suggestion. That's my mistake. The time when I create this repo, it is just a fork of the official one, not an public project. And I do plan to deprecate it after the official one provide better dynamic shape support. I will change the names ASAP.

gengshenghong commented 3 years ago

Thanks @grimoire for your confirmation. You may also need to update the code in your https://github.com/grimoire/mmdetection-to-tensorrt project (not sure whether other projects also reference this library), as it does not work with official torch2rt version as of now. When official one provides dynamic support, you can revert the changes. :)

grimoire commented 3 years ago

Hi I have change the name from torch2trt to torch2trt_dynamic, include mmdetection-to-tensorrt. A lot of files has been changed. If you found any error. Just let me know. Thanks.

gengshenghong commented 3 years ago

Hi @grimoire , I've tested and it worked well to install the same library together now. I'll close this issue.

Thanks, Shenghong