CNN_FineTunning in keras with pretrained imagenet weights

Training deep neural networks from stratch have some problems

1. Take weeks and even months using powerful GPUs in Google or Microsoft Servers.
2. Need huge data set
3. Millions of parameters

To overcome these problems a method called Transfer learning or Fine tuning

Used to train Deep Neural Nets with small dataset, even in CPUs and only a few thousands of parameters will be trained.

We will learn how to use state-of-the-art Deep Learning models to solve a Supervised Image Classification problem using our own datasets with/without GPU acceleration.

Pre-trained Deep Neural Nets trained on the ImageNet challenge are made public and available in Keras.

ImageNet

is a huge image dataset used to help researchers and educators in computer vision track. you can check it from here http://www.image-net.org/

** The main idea based on first or ealier layers extract the general features of any objects ** So we can use this feature and using the depth or last layers to extract the specific features of our objects we want to classify or recognize.

The pre-trained models we will consider are VGG16, VGG19, Inception-v3, Xception, ResNet50, InceptionResNetv2 and MobileNet.

These models for binary classification and currently i will adjust it for multiclassification also.

The code is easy and simple so, you can edit it to run on your owen dataset.

still preparing other models