How to implement a custom plugin equivalent to "tf.tile" in Python API?

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

I tried to convert LPRNET to Tensorrt engine. However some layers such as tf.tile are not supported.
I know that such blocks must be implemented using the API (for me Python API). And have seen the samples, but I am completely confused and not sure where to start.

Any suggestions for:

  • the steps of the procedure, and
  • the implementation details in Python (especially for tf.tile)?
2

Hi @fardo54,
I believe, the below link should be able to help you.

Thanks!

3

Dear @AakankshaS,

Thank you for your reply.

I have already seen this page and the sample codes, but unfortunately I was not able to figure it out.
Some provided samples (in Python) have used the equivalent blocks available in Python API for the unsupported blocks; I could not find some sample code with Python implementation of the unsupported layers from scratch.
Some other sample codes have implemented the custom plugins in CUDA which I am not familiar with…

any suggestion?

Thank you in advance

4

Hi @fardo54,
You can refer to the below link

Thanks!

6

Thanks again @AakankshaS 🙏
I will take a look at that.

7

Hi @fardo54 have you solve this? If yes, would you mind to share? thanks