how can I do object detection with vgg16 by using rcnn??

Hello, I have my custom image data of car and I labeled all my images.

And I want to train CNN by using rcnn. source code is 'rcnn = trainRCNNObjectDetector(gTruth, layers, options)

gTruth is my dataset and I want to use vgg16 for layers.

accodring by this url, https://kr.mathworks.com/help/deeplearning/ref/vgg16.html

I can't modify inputlayers(imagesize) and number of nodes of last classification layer and number of classess.

how can I do object detection with vgg16 by using rcnn??

Answer 1

[Sourav Bairagya]

You can extract the desired number of layers from the pretrained net and define your choice of network. Here is an small example for extracting the desired layers.

net = vgg16;
lastFeatureLayerIdx = 32;
layers = net.Layers;
middlelayers = layers(2:lastFeatureLayerIdx);

Now, you can define your choice of input layer and final layer (i.e. classification layer) and then combine them with the middle layers to get the complete network.

layers = [
    inputLayer
    middleLayers
    finalLayers
    ]

Now, you can leverage this link to get an idea how to perform object detection using R-CNN:

https://www.mathworks.com/help/vision/examples/object-detection-using-deep-learning.html