Skip to content

Implementing Transfer Learning to N>3 Channels Imagery and Removing Pre-processing Layers from Tensorflow.applications

Notifications You must be signed in to change notification settings

phelipedarc/TransferLearning_Muilti-Channel_Imagery

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 

Repository files navigation

TransferLearning_Muilti-Channel_Imagery

Implementing Transfer Learning to N>3 Channels Imagery and Removing Pre-processing Layers from Tensorflow.applications

This is a tutorial Notebook of how to solve two problems:
1.) How to implemente transfer Learning, using the weights of IMAGENET to images with N>3 channels.

2.) How to Remove the peprocessing layers loaded with Tensorflow models

Neste Notebook iremos focar em dois problemas:

1) As redes neurais Convolucionais pré carregadas do tensorflow, com os pesos da imagenet, só aceitam imagens com 3 canais (RGB).

2) As CNN carregadas do TF vêm com camadas de pré processamento embutidas.

Objetivo: Um tutorial de como podemos retirar essa camada de pré processamento, e adaptar a rede para que ela possa ser usada com N canais. Os pesos associados aos canais extras serão a média dos valores dos pesos da Imagenet para os canais originais.

Este código funciona em qualquer rede disponível no TensorFlow.applications 2.8.0

This code was developed by Phelipe Darc.

About

Implementing Transfer Learning to N>3 Channels Imagery and Removing Pre-processing Layers from Tensorflow.applications

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published