CNN and Image Sizes

조회 수: 15 (최근 30일)
Jovan
Jovan 2023년 8월 12일
답변: Josh 2023년 11월 5일
I have a 129x23x2 input and 129x1x2 output for couple of thousand samples in the training set. How do a I create an convolutional autoencoder in matlab to fit the output size? Can somebody explain how do we structure windows of the convolution layers?
  댓글 수: 1
Josh
Josh 2023년 11월 5일
I might need clarification regarding the nature of your images as usually images are represented as a:
  1. n x m matrix where elements values range from [0,1] --- this is more commonly known as an indexed image and can either be grayscale or color if used in combination with a colormap
  2. n x m x 3 matrix where elements range from [0,1] (indexed) or [0,255] --- this generally represents the complete RGB values for the image with separate layers for red (:,:,1), green (:,:,2), and blue (:,:,3).
Your description outlines a 3D output matrix; however, it only has two layers.
For changing the size of output (image size), I would recommend using the imresize() function. This function can be used to either scale an image by a factor or resize images to a desired height (rows) and width (columns). As far as image convolution methods, MATLAB allows for you to specificy which method/kernel to use in the resizing of the image. Again, I may be misinterpreting your question and need more clarificaton about what you mean by "structure windows of the convolution layers".

댓글을 달려면 로그인하십시오.

답변 (1개)

Josh
Josh 2023년 11월 5일
I might need clarification regarding the nature of your images as usually images are represented as a:
  1. n x m matrix where elements values range from [0,1] --- this is more commonly known as an indexed image and can either be grayscale or color if used in combination with a colormap
  2. n x m x 3 matrix where elements range from [0,1] (indexed) or [0,255] --- this generally represents the complete RGB values for the image with separate layers for red (:,:,1), green (:,:,2), and blue (:,:,3).
Your description outlines a 3D output matrix; however, it only has two layers.
For changing the size of output (image size), I would recommend using the imresize() function. This function can be used to either scale an image by a factor or resize images to a desired height (rows) and width (columns). As far as image convolution methods, MATLAB allows for you to specificy which method/kernel to use in the resizing of the image. Again, I may be misinterpreting your question and need more clarificaton about what you mean by "structure windows of the convolution layers".

카테고리

Help CenterFile Exchange에서 Image Data Workflows에 대해 자세히 알아보기

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by