Open ankur219 opened 2 years ago
I think they use tensorrt accelerator in paper. My test in Xavier (using tensorrt accelerate) is as below(Camera is 30fps):
@thinkthinking , how did you manage to deploy it on Xavier (I'm struggling). I converted the wts now I need to build engine. Did you just go to toolkit/deploy and did "cmake ." ?
I got some errors because I did not want to use the ZCam so I commented those lines out of the CmakeList. Then I still get cuda library error
@thinkthinking , how did you manage to deploy it on Xavier (I'm struggling). I converted the wts now I need to build engine. Did you just go to toolkit/deploy and did "cmake ." ?
I got some errors because I did not want to use the ZCam so I commented those lines out of the CmakeList. Then I still get cuda library error
@callmesora Hi, I haven't used author's deploy code directly. I edit the camera part and deploy with ROS. I think there are several tips you need to notice:
CMakeLists.txt
, line 31, it should be cuda not coda;yoloplayer.hpp
, line 19, CLASS_NUM should be 1 not 13, because official end-to-end.pth model can only detect 1 class(car).Otherwise, if you want to only test tensorrt delploy pipeline, I suggest you to build infor_files.cpp
, it's not related to ZCam.
@thinkthinking I have changed all three points above, when I build the infor_files.cpp, the .wts still not transform to .engine successfully. the error is: [E] [TRT] [contatcation] all concat input tensors must have the same dimensions except on the concatenation axis (0), but dimensions mismatched at index 1. Input 0 shape: [128,45,80], Input 1 shape: [128,46,80]
in yololayer.h, the INPUT_H,INPUT_W,IMG_H,IMG_W should be change th what param?
Hi first of all thanks for the code. When I did the inference on GTX 1080 Ti, the speed was 5 frames/sec with a resolution of 640, 384.
In the paper, it has been mentioned that the speed on Titan XP is around 41 frames/sec. Is this speed using TensorRT acceleration or not.
Thanks.