Add TensorRT Support and YOLOv5 #100
Conversation
* [Add] Yolov5 trt HRNet inference on video * Fixed a bug with int overflow on tracking * Revisions before review Co-authored-by: Giannis Pastaltzidis <[email protected]>
|
Hi @stefanopini , just checking in. It has come to my attention that there is a new yolov5 release(version 7). When I tried to convert it to trt with the latest version I think there was an error. version v6.2 worked fine if i remember correctly and it can be loaded by adding :v6.2 in the torch hub str. Anyways, I will have free time and I can handle some tasks before the merge, next week if you would like to. |
|
Hi @gpastal24 , thanks for the update! I've been trying the code on a colab notebook and I should be able to finish working on it soon. I'll try the yolov5 release 7 and switch to the 6.2 if not working. If you have some time this week, I'll ask you to take a look at the code and test the notebook as soon as they're ready, thank you! |
Fix bug in detection locations when tensorrt-converted yolov5 is used
|
Hi @gpastal24 , I managed to test the new code and to make the conversion to tensorrt work with the latest yolov5 too! Could you give it a try too? |
|
Hello @stefanopini , I tried the notebook . It worked great |
|
Amazing, thanks for checking! Going to merge soon |
Add TensorRT Support from #99 (in branch tensorrt) to master branch.