depthai: Requested CNN resources overlaps with RGB camera error on Raspberry + Oak D

Best performance can be achieved w/ 14 shaves/14cmx and 2 NCEs. For that the model needs to be compiled for 7 shaves (7 shave for each NCE = 7x2=14) In depthai.py the following config needs to be changed to 14, 14, 2.

if default_blob:
    #default
    shave_nr = 7
    cmx_slices = 7
    NCE_nr = 1

It will likely exceed 30 fps, so the rgb camera fps needs to be adjusted to 40 fps (or more): -rgbf 40 If inference is requested on both left and right cameras 2 NCEs are mandatory, otherwise they won’t be balanced.

@nullbyte91

I found that root cause: left camera socket was ejected. I reconnect camera and the first device works good now.

@nullbyte91 Mobilenet SSD v2 works nicely. Please see this notebook: https://colab.research.google.com/github/luxonis/depthai-ml-training/blob/master/colab-notebooks/Easy_Object_Detection_With_Custom_Data_Demo_Training.ipynb

@iamsg-luxonis Yes. I’m using the latest version. I will try with 2020.1/2020.2 and update you. Thanks again for the support. One quick question about precision support, Can we able to use both FP16 and U8 format?

Thanks @asmorkalov for the error report. Very interesting and we’ll check it out soon.

CC: @iamsg95 and @themarpe