Pad layer only works in one dimension in TensorRT5

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

Consider the following tensorflow code:

placeholder_tensor =  tf.placeholder(dtype=tf.float32, shape=(1, 361, 641, 3))
image_tensor_tensor = padmap(placeholder_tensor, 3 )

def padmap(x, size):
    paddings = tf.constant([[0, 0], ,
                                  , [0, 0]])
    x = tf.pad(x, paddings, "CONSTANT")
    return x

The expected output is a padded tensor
<tf.Tensor ‘Pad:0’ shape=(1, 367, 647, 3) dtype=float32>

TensoRT issue
i converted this code via uff and now in tensorrt 5.0.4.3 ouptut size is as below

Placeholder x1x361x641x3
Pad x1x361x647x3

its padded only in one direction

But Tensorrt document clearly says following-

Conditions And Limitations
‣ A must have three dimensions or more.
‣ The padding can only be applied along the two innermost dimensions.
‣ Only zero-padding is supported.

Configurations: Windows, TensorRT 5.0.4.3

2

I wish you’d get an official reply, but if I’d hazard a guess it’s probably because “The padding can only be applied along the two innermost dimensions” and the fact that you’re applying padding on the 3rd-last and 2nd-last dims instead.