Change number of output classes in Squeezenet?
16 visualizaciones (últimos 30 días)
Mostrar comentarios más antiguos
I want to retrain a slightly modified squeezenet on the CIFAR100 dataset. Here are the last few layers of Squeezenet:
Usually, a fully connected layer is replaced to change the number of output classes, or the pooling layer is changed. However, MATLAB's Deep Network Designer doesn't give any options to adjust output size of the globalAveragePooling2dLayer. And Squeezenet doesn't have any fully connected layers.
My question is, how do I change the network output class size from 1000 (imagenet) to 100 (cifar100)?
1 comentario
Manav Madan
el 22 de Jun. de 2022
First store the values set for different features in last 4 layer the "convolution2dLayer" then delete and select a new "convolution2dLayer" where copy rest of the values for parameters and change "NumFilters" to the number of classes you have.
Respuestas (0)
Ver también
Categorías
Más información sobre Image Data Workflows en Help Center y File Exchange.
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!