Train a Siamese Network to Compare Images with image size [227 227 3]

10 次查看（过去 30 天）
MAHMOUD EID
MAHMOUD EID 2023-4-17
评论： MAHMOUD EID 2023-4-19
Hi,
I have followed the official tutorial to train Siamese Network to Compare Images
link: https://se.mathworks.com/help/deeplearning/ug/train-a-siamese-network-to-compare-images.html
The trained image of the network has a size of [105 105 1]. when I am trying to train a network on images with size [227 227 3] , I usually recieve many errors.
Can someone help which parts in the code need to be change?

请先登录，再回答此问题。

回答（1 个）

Sam Chak
Sam Chak 2023-4-17
You should be able to find this line in the tutorial:
imageInputLayer([105 105 1],Normalization="none")
  1 个评论
MAHMOUD EID
MAHMOUD EID 2023-4-19
Hi sam,
Thanks for support.
I have done so on my image dataset and figured out where is the error. Usually the code runs only on the CPU.
-When choosing 'Auto' or 'GPU' , usually the code stop due to unsuffiecient memory in the GPU . I have tried to reduce the miniBatchSize to the minimuim number for example 4 or 5 , and still facing the same error. Do you have an idea tackling this isssue?
- the second point: During training, I notice that the loss during training is usually a fixed number (0.7) and not change across the 1000 iterations. I have tried to change learningRate = 6e-5;
gradDecay = 0.9;
gradDecaySq = 0.99;
but the losses during training not changing, what is supposed to change to reduce the losses during training?
@Sam Chak

请先登录，再进行评论。

类别

在 Help Center 和 File Exchange 中查找有关 Image Data Workflows 的更多信息

标签

Community Treasure Hunt

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

Start Hunting!

Translated by