BAD RESULT

[GaHooooo]

Hi, thaks for your great work! but i have some problem in this work

This is my device GPU:GTX1080Ti CUDA : 10.2 cudnn :8.2 TensorRT:7.1.3.4

but the result is bad i found that the time used almost increase hundredfold than your report（PostProcessDuration Time） [11/18/2021-10:12:50] [I] [INFO] pointNum : 278272 [11/18/2021-10:12:51] [I] PreProcess Time: 49.6466 ms [11/18/2021-10:12:51] [I] inferenceDuration Time: 28.344 ms [11/18/2021-10:12:51] [I] PostProcessDuration Time: 124.939 m

0a0d6b8c2e884134a3b48df43d54c36a.bin.txt

and i open the result found the ouput has 1445 point i follow your readme https://github.com/CarkusL/CenterPoint/tree/main/tensorrt/samples/centerpoint

CAN YOU HELP ME

[CarkusL]

Did you use -O3 flag when you compiled the program? https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

[CarkusL]

By the way, The post-processing runs on the CPU, and it is not accelerated with GPU. It will take a lot of time, when the model outputs many boxs

[GaHooooo]

Did you use -O3 flag when you compiled the program?

https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

i use -O3 flag My biggest confusion is that why my result is different from yours，my result has 1445 boxes

[CarkusL]

Did you use -O3 flag when you compiled the program? https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

i use -O3 flag My biggest confusion is that why my result is different from yours，my result has 1445 boxes

why does your result have so many boxs?

[ZHUANG-JLU]

Did you use -O3 flag when you compiled the program? https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

i use -O3 flag My biggest confusion is that why my result is different from yours，my result has 1445 boxes

I met the same problem,1445 boxes

[aisaack]

Did you use -O3 flag when you compiled the program? https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

i use -O3 flag My biggest confusion is that why my result is different from yours，my result has 1445 boxes

I met the same problem,1445 boxes

I'm facing same problem. It seems the command line ./centerpoint automatically apply fp16 quantization. How to avoid applying it?

[aisaack]

Did you use -O3 flag when you compiled the program? https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/Makefile.config#L263

i use -O3 flag My biggest confusion is that why my result is different from yours，my result has 1445 boxes

I met the same problem,1445 boxes

I'm facing same problem. It seems the command line ./centerpoint automatically apply fp16 quantization. How to avoid applying it?

Since the sample doesn't use fp16 computation and I don't have fp16 supported GPUs, the code line should be changed from https://github.com/CarkusL/CenterPoint/blob/9980cdb56fb928528e69937f415071613eedf028/tensorrt/samples/centerpoint/samplecenterpoint.cpp#L370

to

     params.fp16 = args.fp16; 

or to

    params.fp16 = false;

Then similar result can be achieved.