Introduction to CNN Exam 19Questions with Verified Answers
Convolution CORRECT ANSWER - is a specialized linear operation on an image (a function), used for feature extraction (edges)
The linear operation uses a sm
...
Introduction to CNN Exam 19Questions with Verified Answers
Convolution CORRECT ANSWER - is a specialized linear operation on an image (a function), used for feature extraction (edges)
The linear operation uses a small array of numbers called a CORRECT ANSWER - filter/ kernel
The operation of filter over the image is an CORRECT ANSWER - pixel-wise product
The resulting products are summed to obtain the output value in the corresponding pixel position of the output image, also called a CORRECT ANSWER - feature map
Convolution formal definition CORRECT ANSWER - is repeated applying multiple kernels to form an arbitrary number of feature maps,
Each feature maps represent CORRECT ANSWER - different characteristics of the input image. Thus the different kernels can be considered as different feature extractors
Two key hyperparameters that define the convolution operation are CORRECT ANSWER - size and number of kernels
The size of kernel is typically CORRECT ANSWER - 3 × 3, but sometimes 5 × 5 or 7 × 7
.The number of kernels determines the CORRECT ANSWER - depth of output feature maps
The filters are shared among all the neurons of the layer. This is known as shared weights and bias. CORRECT ANSWER -
Size of the new image in row and column is given by CORRECT ANSWER - Rnew = ((image_rows - filter_rows) / stride) + 1. Similarly, Cnew = ((image_cols - filter_cols)/stride) + 1
Filter convolution increases output image size. CORRECT ANSWER - False
A technique called padding is used to prevent image size reduction CORRECT ANSWER - True
When we need large depth i.e. number of convolutional layers in the network, we need to pad the regions appropriately else we loose edge information in successive layers CORRECT ANSWER - True
The filters are shared among all the neurons of the layer. This is known as shared weights and bias CORRECT ANSWER - True
Two key hyperparameters that define the convolution operation are size and number of kernels. CORRECT ANSWER - True
.The size of kernel is typically 3 × 3, but sometimes 5 × 5 or 7 × 7. CORRECT ANSWER - True
The number of kernels determines the depth of output feature maps. CORRECT ANSWER - True
What does this tell us about images ? CORRECT ANSWER - 1. For object recognition we need high level features.
2. Many of the low level features are redundant
3. In this example, many of the pixels are unimportant
4. Only pixels that for the edge of the object are important and that too no all pixels on the edge
1. Raw pixel values to intensity gradients
2. Gradients point in the direction of max increase in pixel intensity
3. The entire image can be considered as a vector field of image gradients
[Show More]