is the experiment not correspond your paper?have onle 2% save time?

[liaohengrui]

VTM11 run basketballdrill qp22 50poc result

your code in cpu，other config is same,i do a statistic the cnn use 30s in cpu

[liaohengrui]

because i use cpu run cnn ,so user time is fault . only see elapsed is ok。2757 -2689 / 2757 = 2% ？otherwise 2757 -(2689 - 30（cpu for cnn）)/ 2757 = 3.5%?

[unacanal]

Hi! Thanks for your interest in my research.
The experimental results presented in the paper are the results of inference using GPU. Thanks. As you can see here, all script files are run on the GPU.

[liaohengrui]

Hi! Thanks for your interest in my research. The experimental results presented in the paper are the results of inference using GPU. Thanks. As you can see here, all script files are run on the GPU.

Thank you for your reply. I understand your point. The experiment was originally conducted on a GPU, but due to certain reasons, I couldn't use the GPU for execution. As a result, I measured the execution time of the CNN on a CPU in order to compensate for the time difference between GPU and CPU execution. When calculating the time, I subtracted the total CPU execution time for the CNN from the total time: 2757 - (2689 - 30 (CPU for CNN)) / 2757 = 3.5%. I would like to confirm if my result is accurate. Could it be possible that the result is underestimated due to specific tasks and hardware configurations? Can you help test the configuration "basketballdrill -q 22 -f 50" for me?