Unable to configure ISlice Layer for padding

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

Description

Unable to use ISlice layer for padding a tensor

I am trying to use the ISlice Layer with Tensorrt v8.2.1, since IPaddingLayer is going to be deprecated. I am trying to convert a tensor of shape 1,3,10,10(N,C,H,W) to 1,3,14,14(N,C,H,W) with padding done to top, bottom ,left and right of the tensor. I am trying to use the below code snippet but unsuccesful

ISliceLayer* slice = network->addSlice(*InputLayerITensor,
Dims4{0,0,0,0},
Dims4{1,3,14,14},
Dims4{1,1,-2,-2});
I see error

Error occured when invoking TensorRT API

4: (Unnamed Layer* 0) [Slice]: out of bounds slice, input dimensions = [1,3,10,10], start = [0,0,0,0], size = [1,3,14,14], stride = [1,1,-2,-2].

Error occured when invoking TensorRT API

4: [network.cpp::validate::2871] Error Code 4: Internal Error (Layer (Unnamed Layer* 0) [Slice] failed validation)

Any help in usage of this API for my requirements is highly appreciated.

Environment

TensorRT Version: Tensorrt v 8.2.1
GPU Type: Titan xp
Nvidia Driver Version: cuda 11.2
CUDA Version: cuda 11.2
CUDNN Version: 8.2.1
Operating System + Version: Linux
Python Version (if applicable):
TensorFlow Version (if applicable):
PyTorch Version (if applicable):
Baremetal or Container (if container which image + tag):

Relevant Files

Steps To Reproduce

Build Makefile using the executable. attached in the zip and run the exe.

Please include:
Build Makefile using the executable. attached in the zip and run the exe.

#2

Hi,
Please check the below link, as they might answer your concerns

Thanks!

#4

Hi,

Thanks for pointing me to the documentation.
By going through API documentation for addPadding I came to know that it will be removed and I need to use addSlice.

I found the below example in API documentation(TensorRT: nvinfer1::ISliceLayer Class Reference), it is not clear and does not mention how to use it for padding.

For example using slice on a tensor: input = {{0, 2, 4}, {1, 3, 5}} start = {1, 0} size = {1, 2} stride = {1, 2} output = {{1, 5}}

When the sliceMode is kCLAMP or kREFLECT, for each input dimension, if its size is 0 then the corresponding output dimension must be 0 too.

A slice layer can produce a shape tensor if the following conditions are met:

** start, size, and stride are build time constants, either as static Dims, or computable by constant folding.*
** The number of elements in the output tensor does not exceed 2Dims::MAX_DIMS.

The input tensor is a shape tensor if the output is a shape tensor.

It will be very helpful for us, if there are shipped examples on how to use each layer for specific purposes.

Thanks,
Praveen.

#5

Hi,

We support onnx pad using iSliceLayer in parser as following.

Thank you.