Part Number: TDA2
Hello,
I have gone through most references available for TIDL on TDA2x. The TIDL Library User Guide and Datasheet are nice starting points to understand the deployment of neural networks on TDA2. But the deployment procedure described in the Guide works in a disjoint fashion, i.e., for a neural network whose layers are split between EVE and DSP cores, we have to first execute the layers on EVE, get the output, and then feed it to execute the layers on DSP. Obviously, this cannot happen in a practical scenario; I will have to set up the entire network in one go.
I also went through the TIDL Library API described at this link. But this whole chapter appears to apply to TIDL for AMxx processors, and I did not find any equivalent reference for TDA2x processors.
Can you point me to some sample applications which cover an end-to-end scenario, e.g., test image(s) fed to a neural network which is split across EVE and DSP, where the interlinking and I/O of the EVE and DSP is handled in real-time by the program, and then visualization of the final output?
Thank you.
