It is technique which we can use the pre-trained weight to learn our new data. The main advantages over the pre-trained weights are, those models were already trained on various data so it is easier for us to make the model learn in our new data. To train the model from the scratch, we would be needed high end servers and it will take much larger time to train on.
There are many models are available in imagenet like VGG16, VGG19, ResNet, and so on. In this project I used VGG19.
As I mentioned above, that model is already trained on globally available objects. So the aim is to pass the image and try to get the object correctly.
Since this is my first project, I did not go for any further training, it is just the prediction of the model.