Training a FasterRCNN Network?

Hi,

I am trying to train a fasterRcnnNetwork using a pretrianed 'Alexnet'. According to documentation for that, I have to use the trainFasterRCNNObjectDetector command.

My Question is do I have to modify the network as described on this link; https://www.mathworks.com/help/vision/ug/faster-r-cnn-examples.html#mw_51a5e174-51f2-4da8-892b-16d3ce2278e2 or if I give it the trained network name, it will do it automatically?

Thanks

[Kevin Chng]

Hi,

The example is to build deep learning model from scratch.

If you want to use pretrained network with alexnet, first of all, do you want to change the classification layer? or maintain the original 1000 categories in alexnet.

if you have your own classification/categories, you may need to do transfer-learning first.

[numi khnax]

My problem is actually about object detection and I know that the pretrained alexnet available in matlab is used for image classification. I would like to create a fasterRCNN network for object detection and in Matlab this can be accomplished using this command: trainFasterRCNNObjectDetector. Also, I am not going to use it to detect 1000 different objects, I only need to detect one type of objects. So I know if I am doing image classification I have to manually modify the classification layers before doing transfer Learning.

For object detection we don’t only modify the existing layers but we also add other layered such as ROI pooling etc. So do i have to manually add those layers manually (as shown in the link provided up in my question)or the command will do it automatically??