I run the code for HO3D locally, following the same command here(python run_ho3d.py) in README. My desktop has a RTX 3090 GPU (which is the same as the device mentioned in the paper) and 12 Intel i5-12400 CPUs. I test the inference speed by timing every iteration of this for loop. The speed on my desktop is around $0.4Hz$, much slower than $10.2Hz$ reported in the paper. Why is it so slow on my desktop? Do I time inference time correctly? Or is there any hyperparameter that I need to change?
wenbowen123
commented
7 months ago
Thanks for the amazing work.
I run the code for HO3D locally, following the same command here(
python run_ho3d.py) in README. My desktop has a RTX 3090 GPU (which is the same as the device mentioned in the paper) and 12 Intel i5-12400 CPUs. I test the inference speed by timing every iteration of this for loop. The speed on my desktop is around $0.4Hz$, much slower than $10.2Hz$ reported in the paper. Why is it so slow on my desktop? Do I time inference time correctly? Or is there any hyperparameter that I need to change?Thanks in advance.