Reshape tensor [1,640,640,3] to [1,3,640,640]?

Research & Models
, , , ,
#1

Hello,
Is there a way to reorder the shape of the tensor? Here’s a code sandbox, its in predictWebcam function
https://codesandbox.io/p/sandbox/patient-brook-z6s3w9

I’m getting this error:
Uncaught Error: The shape of dict[‘images’] provided in model.execute(dict) must be [1,3,640,640], but was [1,640,640,3]

Thank you, greatly appreciate the help!

#2

Hi @Bass_Guitarist ,

Could you please try below code and see if it works.

You can use the tf.transpose() function.

To use the tf.transpose() function, you need to pass in a list of integers that represent the new order of the dimensions. For example, to reorder the dimensions of the tensor from [1,640,640,3] to [1,3,640,640], you would pass in the list [0,3,1,2].

# Reorder the shape of the tensor to match the model's input shape.
image = tf.transpose(image, [0, 3, 1, 2])

For more detailes refer this TensorFlow Documentation.

I hope this will help you!

Thanks

#3

Thanks Laxma_Reddy_Patlolla. That worked. I used tf.transpose(tensorShape.expandDims(), [0, 3, 1, 2]);

1 Like