How to handle images of different sizes that are smaller than the input layer of a deep learning model?

Question

I am performing human awareness detection and have trained my model using transfer learning with MobileNetV2. This model expects a tensor of dimension [Null,224,224,3].

I have applied face detection using BlazeFace which uses an input of [128,128,3] on the input video stream and cropped the detected faces in order to send the cropped faces to my custom model but I am not sure what to do as the cropped images are all of varying sizes and smaller than what my model expects.

Example of a cropped face tensor

Array [
  1,
  43,
  111,
  3,
]

Possible duplicate https://datascience.stackexchange.com/questions/30819/image-resizing-and-padding-for-cnn — YuseqYaseq, Sep 16 '20 at 15:08

score 0 · Accepted Answer · answered Sep 17 '20 at 04:46

0

The issue was fixed by resizing the Tensor to fit into the model. I was reshaping them instead of resizing them.

answered Sep 17 '20 at 04:46

yudhiesh

213
1
9

how did you resize the Tensor? – Syed Ali Nov 20 '20 at 01:00
If you are using tensorflow then tf.image.resize() would be the way to do so. – yudhiesh Nov 20 '20 at 02:13

How to handle images of different sizes that are smaller than the input layer of a deep learning model?

1 Answers1