Project 6 Mini Project
Mobile Net V2 is a pre-trained model for image classification. pre-trained models are deep-trained models that are trained using a large image dataset.
- Mobilenet simplifies the process of image processing. Image processing helps transform the image dataset into a format the model can understand to give more accurate results.
- Mobile_net significantly reduces the number of parameters thus making it less complex.
In pointwise convolution either kept the number of channels the same or doubled them. In v2 it does the opposite, it makes the channels smaller. This is why known as the projection layer. It projects data with a high number of dimensions into a tensor.
For example, the depthwise layer may work on a tensor with 144 channels, and the projection layer will then shrink down to 24 channels. This kind of layer is called the Bottleneck layer it reduces the amount of data that flows through the network.
The first layer is the new kind in the block. This is also a 1*1 convolution. Its purpose is to expand the number of channels in the data before it goes into the depthwise convolution. Hence, this expansion layer always has more output channels than input channels. Exactly how the data gets expanded is given by the expansion factor. So the output and input of the block are low-dimensional tensors, while the filtering is done on a high-dimensional tensor.
The second new thing in mobile net v2 is a building block is a residual connection. This works just like in resNset and exists to help with the flow of gradients through the networks.
