mrrahul011 / eip_project

EIP_Project

Result

print(score) = [0.02607954332709196, 0.9925]

#Definition

1. Convolution:
   Convolution is the process of detecting various patterns from an image using filters/kernals. Element wise multiplication of pixel in a kernel when super imposed on an input image and then summing it up to get a single value out. This is to extract the dominating feature .

2.Filters/Kernals:
Filters/Kernals also known as feature extractor is gener- ally a 3 x 3 matrix that convolve over an input image to extract certain feature from the input image.

3.Epochs: The number of times a network look through the entire image data set. If a network looks the entire image data set once, its called 1 Epochs.

4.1X1 Convolution: The 1x1 convolution combines the channel which are contextually linked together.Its used to reduce the depth of a layer with out loosing much information.

5.3x3 Convolution: A 3x3 matrix is used to convolve over the image ie 9 pixel get multiplied with corresponding pixel in input image and gets added up to give a single value. Generally, After this convolution 2 pixel are reduced in X and Y each.

6.Feature Maps: Feature map is the result of convolution after applying activation function.

7.Activation Function: Activation function are used to make the weights of a kernel non linear.

8.Receptive Field: The number of pixel in an input image that the filter can look at, ie a 3x3 filter can look at 9 pixel on an input image. This is local receptive feld ie size of kernal.