CIFAR-10 object detection – revisited
In the previous chapter, we trained a simple convolution neural network (CNN) model on the CIFAR-10 dataset. Here, we are going to demonstrate the case of using a pre-trained model as a feature extractor while removing the fully connected layer of the pre-trained model, and then we'll feed these extracted features or transferred values to a softmax layer.
The pre-trained model in this implementation will be the inception model, which will be pre-trained on ImageNet. But bear in mind that this implementation builds on the previous two chapters that introduced CNN.
Solution outline
Again, we are going to replace the final fully connected layer of the pre-trained inception model and then use the rest of the inception model as a feature extractor. So, we first feed our raw images in the inception model, which will extract the features from them and then output our so-called transfer values.
After getting the transfer values of the extracted features from the...