Hi community, first, I am glad to have found this forum as it may well be my last chance of resolving some issues that I wish to resolve. Last year (Aug 2022 ) I used the ‘Figurine.Android’ colab notebook to create a number of custom .tflite models to work on my Raspberrypi. These all worked well - thanks to Khanh LeViet’s YouTube video and his colab notebook. Last week, I followed his last(?) video ‘Make object detection faster by using Coral’. The .edgetpu models that I downloaded, sadly, do not work. The Coral TPU workd with other available datasets, so there isn’t anything wrong with the hardware. The terminal outputs the following, ‘Segmentation error’. Has anyone succeeded in achieving the re-use of their existing 'tflite models and if so any help in achieving this would be most welcome.
I had had similar problem. When I tried strace it turned out the erroring goes while memory allocation. In short - not enough available RAM. BTW, thanks for these Khanh LeViet’ videos. Much appreciated.
Thanks for taking the effort to reply. BTW I have found that it is possible to obtain 30fps on a live video stream using the MediaPipe Object Detection Model WITHOUT the Coral TPU! This was on a Raspberry Pi 4 with 4GB RAM. I have now obtained a RPI 5 with 8GB ram and will be testing my use case model out as soon as I have time to set up all the hadrware and software. However, the greatest task I have is to document everything so that I can repeat my results - all this takes a lot of time.