Closed thelateperseus closed 2 years ago
@thelateperseus Thanks for tagging me on this. I'll have to take a look and test it out.
Increasing the size of the swap space makes sense. We usually recommend that users disable the GUI to free up enough memory to get that example working. My only hesitation with this method, is that it increases the number of read/write cycles to the SD card and therefore, it also increases the chances of corrupting the SD card.
If possible, maybe an alternative would be to modify the swap file size only while the example is running. Though, I'd have to read up more on Docker containers.
I am using the SparkFun JetBot AI Kit v3.0 (Without Jetson Nano) with a Jetson Nano 2GB I bought separately. I downloaded the sparkfun_2GB_jetbot_v00-02_linux.zip image and wrote this to the provided SD card. I followed the SparkFun assembly and setup guide and got the basic motion and teleoperation examples working without issue.
However, the camera freezes when inference starts in the collision avoidance "live_demo.ipynb" example. I found a related thread on the Nvidia forum Rpi v2 CSI camera freezes with Jetson Nano 2GB, and it seems that this is a common problem with the 2GB Nano. I fixed the issue by adding 6GB swap space, reducing other memory usage, then allocating 8G swap space to the jetbot_jupyter container:
Given that this is a common issue with 2GB Nanos with CSI cameras, I think that the default Jetbot configuration should set 8G of swap space on the jetbot_jupyter container instead of 3G.
@santaimpersonator fyi in case you want to update the SparkFun image.