Transfer Learning

Transfer Learning of Pre-trained Neural Network or Imported ONNX Classification Model in GUI
14.7K Downloads
Updated 5 Nov 2019

View License

It allows user to do transfer learning of pre-trained neural network, imported ONNX classification model or imported MAT file classification model in GUI without coding. Other than using the existing model, user can design their neural network using Deep Network Designer (MATLAB built-in application) and later use this app to train the neural network. User can analyze the pre-trained network or imported network in this app. It also allows user to modify the image augmenter information and training option (hyperparameters). Once completed the training of network, user may export the trained network to workspace, MAT file, ONNX file and generate MATLAB code for the steps being done in the application. User can generate NVIDIA GPU CUDA Code, however it is not supporting imported ONNX model.

Pre-trained Neural Network available :
1) alexnet
2) googlenet(ImageNet)
3) goolgenet(Places365)
4) resnet18
5) resnet50
6) resnet101
7) vgg16
8) vgg19
9) inceptionv3
10) inceptionresnetv2
11) squeezenet
12) densenet201
13) mobilenetv2
14) shufflenet
16) xception
17) nasnetmobile
18) nasnetlarge
19) import ONNX model
20) import mat file model

[Up to R2019a, imported ONNX layer is not supporting generation of CUDA code]

How to design & train neural network using this app?
https://www.youtube.com/watch?v=-GeZa6IL2QA

How to import ONNX Model to train using this app?
https://www.youtube.com/watch?v=8pZiduqp35g

Cite As

Kevin Chng (2024). Transfer Learning (https://www.mathworks.com/matlabcentral/fileexchange/70293-transfer-learning), MATLAB Central File Exchange. Retrieved .

MATLAB Release Compatibility
Created with R2019b
Compatible with any release
Platform Compatibility
Windows macOS Linux

Community Treasure Hunt

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

Start Hunting!
Version Published Release Notes
2019.6.2

Debug

2019.6.1

Debug MATLAB Code Generation

2019.5.1

Other than use sub-folder's name as label source, now you may import your MATLAB workspace variable (vector) as label source for your image. For future development, this app is looking to expand its capability to deep learning for regression.

5.3

Update description

5.2

Fix bug

5.1

Change Description

5.0

Allow user to import customized model designed from MATLAB Deep Network Designer

4.4

Change description

4.3

* Add video to demonstrate how to import ONNX model for this application.

4.2

Missing Picture

4.1

*change description

4.0

New Feature : Allow to import ONNX Classification Model for transfer learning, however, the imported ONNX model is not supported to generate GPU CUDA Code in this app.

3.4

Add more pre-trained neural network : xception,shufflent,nasnetmobile,nasnetlarge

3.3

Small Update (delete old result when start new session)

3.2

New Feature : Allow to change the learning rate of new last learnable layer

3.1

fix bug

3.0

Add new features:
1) Additional pre-trained neural networks : googlenet (Places 365), densenet201,mobilenetv2
2) More information about pre-trained neural network in display : Size, Number of Parameters (Millions)
3) Generate NVIDIA GPU CUDA Code

2.0

Change GUI

1.06

Change Title

1.05

Add the link of documentation

1.04

Update

1.03

*Add documentation to explain the apps

1.02

Change Description

1.1

*Change Description

1.0