HI I test my own custom data with 12 classes on yolov4 and export it to yolo4-FP32.rt file. While I converted I observe this error in debugging
====== CUDNN inference ======
Data dim: 1 3 416 416 1
Data dim: 1 51 13 13 1
===== TENSORRT inference ====
Data dim: 1 3 416 416 1
Data dim: 1 51 13 13 1
=== OUTPUT 0 CHECK RESULTS ==
Error reading file yolo4/debug/layer139_out.bin with n of float: 551616 seek: 0 size: 2206464
/home/xavier/Downloads/tkDNN/src/utils.cpp:58
Aborting...
but it generates .rt file when I test it on video. It returns me these results with batch size 1.
Time stats:
Min: 59.4498 ms
Max: 215.051 ms
Avg: 75.7361 ms 13.2037 FPS
and for FP16 Results with batch size 1 are:
Time stats:
Min: 27.9283 ms
Max: 88.9858 ms
Avg: 34.4137 ms 29.0582 FPS
Now here I want to confirm a few things:
1st: FPS is low as reported in the repo.
2nd: How FPS is calculated here and what is your criteria/ Normally in python, we calculate it as 1 / start_time - end_time
between inference statements (Can u PLease guide).
@ceccocats @mive93
Thanks
HI I test my own custom data with 12 classes on yolov4 and export it to yolo4-FP32.rt file. While I converted I observe this error in debugging
but it generates .rt file when I test it on video. It returns me these results with batch size 1.
and for FP16 Results with batch size 1 are:
Now here I want to confirm a few things: 1st: FPS is low as reported in the repo. 2nd: How FPS is calculated here and what is your criteria/ Normally in python, we calculate it as
1 / start_time - end_time
between inference statements (Can u PLease guide). @ceccocats @mive93 Thanks