Input format is RGBA by config or by default, but input channel larger than 4

edited October 2021 in AI Model migration

Hello,

I was trying to run a complicated model on Kneron dongle, which had a lot of unsupported layers in the middle. I decided to split the model in the middle and run the two parts separately.

However, when I try to convert the second model (tflite), which uses feature map output from first model as input, the converter gives an error

 Hardware not support: Input format is RGBA by config or by default, but input channel larger than 4

The input is a feature map with shape (1, 56, 56, 256).

Is the hardware not support channels more than 4 or it was thinking the input is an image?

Can I configure it to accept feature map input?

I'm using the python docker toolchain.

Thanks

Comments

  • Our solution currently does not support channels more than 4 .

    You can provide information about your unsupported layers for us to see how we can help you

    Thanks!

The discussion has been closed due to inactivity. To continue with the topic, please feel free to post a new discussion.