Change input size of a pre-trained network
45 views (last 30 days)
I am working on an object detection algorithm with YOLO V2 and I have been following Mathworks guidelines. In particular, I wanted to use the solution given in the following link: https://uk.mathworks.com/help/vision/ug/create-yolo-v2-object-detection-network.html. After that, I'd like to re-train the network so that it gets used to the type of images I am working with. However, when it comes to modify the input size, I end up by having a graph structure, and my imported network continues to have the same input size. If I try to modify it manually, it says the InputSize property is a read-only property, and if I try to directly change my new imageInputLayer object in the network, it also says it's read only.
Is there a way to:
- Import a pre-trained network (I am using resnet50)
- Change its input size and number of output features
- Re-train it?
Thank you in advance!
HyeongHun LEE on 11 Jun 2019
Edited: HyeongHun LEE on 11 Jun 2019
If you want to change the specific layer parameters in pretrained neural networks(e.g. ResNet, DenseNet etc), following the procedure will work.
1. Load target pretrained network in workspace
2. Open "Neural network designer (GUI version, newly updated in 2019a)"
3. Import pretrained network model into the neural network designer space (block diagram will display automatically)
4. Change layer properties (eg. input size, filter size etc)
5. Export network model