Combine customized neural network with CNN
5 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
As I found we can customized each layers in a CNN, and we can also do customizing our own neural network by "net=network" and set properties of "net". But can those class of CNN layer be used in a general neural network? I want to combine the output of two CNN as two channel of 'image' and do CNN on this new 'image'
0 comentarios
Respuestas (1)
Chaitral Date
el 24 de Abr. de 2017
Regarding your first question, No, you shouldn't be able to use CNN layer classes in a general neural network. It'll still take on somewhat a similar form to a CNN. There is some flexibility, but a lot of limitations at the same time.
Regarding your second question, The two channels portion is very unclear. What is the output that you are currently getting from the two CNN?
3 comentarios
Jason Vann
el 1 de Ag. de 2017
This is exactly the kind of feature I would like to see. I would like to develop a network with parallel CNNs and connected at the end with a fully connected and classification layer. Looking at the R2017b pre-release, there is a custom layer capability which may help address this issue, but it would be a very convoluted work-around.
Alexander Tarroni
el 15 de Sept. de 2017
I am also trying to fuse two pre-trained VGG16 models with a single softmax layer at their output, with two inputs taking separate data. It seems you can't make convolutional layers in the custom neural network functions, is there any way to fuse layers from different convnets otherwise?
Ver también
Categorías
Más información sobre Image Data Workflows en Help Center y File Exchange.
Productos
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!