Hi @usuario1918, the jetson_clocks script will lock their clocks to their maximum values as defined by the current power mode (in your case, 5W mode). Typically there is the Dynamic Voltage & Frequency Scaling (DVFS) governor which adapts the clocks based on workload (jetson_clocks disables this). Once detectnet.py is up & running though, the clocks should already be pretty stable.
In that YouTube video, Jetson Nano was running in 10W mode, which gives more performance. If you require higher FPS, are you able to change from 5W to 10W mode?
Did the window close, or did the system turnoff? My guess is that your power supply wasn’t capable of supplying the load needed for 10W mode. Running jetson_clocks may make that more immediately obvious, since it will draw more power all the time (as opposed to only under considerable workloads) If it continues to be an issue, you may want to look into upgrading your power supply.
Hi @usuario1918, you should run it after each reboot if you want it to remain active. Although changing the nvpmodel to 10W mode will be the main thing that improves your performance with jetson-inference. Running detectnet/detectnet.py on a continuous video stream or camera on Jetson Nano will result in the clocks being maximized by the dynamic load governor anyways.