bleh Flashcards
(T/F)Increasing the number of samples within a fixed range results in higher resolution.
True
A grayscale image has three different color channels for every pixel.
False
The morphological opening of an image is simply a dilation followed by an erosion.
False
Watershed algorithm is used to smoothing the edges in the image.
False
Canny edge detection uses alignment of the edges when detecting the edges.
False
Dilation operation in morphological filters can be used removing small noisy pixels from the input image.
True
In deep learning pipeline, hand-crafted features are used for classification.
False
If we build a neural network with no activation function, we get a linear classifier.
True
From the following list of image enhancement techniques covered in class, select the techniques to reduce noise in the original image.
- Median Filtering
- Weighted Average
From the following list of image enhancement techniques covered in class, select the techniques to sharpen the edges in the original image.
- Laplacian
- Sobel Filter
- Unsharp Masking
From the following list of image enhancement techniques covered in class, select the techniques to improve the contrast in the original image.
- Contrast Stretching
- Histogram Equalization
- Power Law
- In an input image, each pixel is shown with 8 bits. In order to find the type of the noise, a test pattern is cropped from a flat region in the input images and its histogram is shown. The histogram of the test pattern is shown below. What kind of noise can be estimated from this histogram?
Salt and Pepper
For ImageNet classification challenge, several convolutional neural networks are designed over the years to achieve better performance. Please sort the following networks from the shallowest to the deepest (from less number of layer to more number of layers).
VGG
AlexNet
ResNet
In our class, we covered four different machine vision tasks using Convolutional Neural Networks.
Which task detects all objects in the image and identifies the pixels that belong to each object?
- Instance Segmentation
Which components of a Convolutional Neural Network is used to reduce the dimensions of the feature maps? Thus, it reduces the number of parameters to learn and the amount of computation performed in the network.
Pooling
When we visualize the first layer of a deep network, what will we see in the output of these layers?
Low Level features such as edges and corners
Fill In the blank in following sentence: Generative Adversarial Networks (GANs) jointly train generator and ___________ with a minimax game.
Discriminator
Select the answers in order to fill in the blanks in the following sentence:
During the backward pass, each node in the graph receives ____________ gradients and multiplies them by ____________ gradients to compute ____________ gradients.
Upstream, local, downstream
Select the answers in order to fill in the blanks in the following sentence:
In the stereo cameras, we search over disparity to find the correspondence in two stereo images. The disparity of the close objects in the stereo images ___________.
Is larger than far objects
Which of the following best describes optical flow in computer vison?
An algorithm for estimating the motion of objects in a visual scene
Which challenge is commonly encountered when estimating optical flow in dynamic scenes?
Illumination changes
What is the primary purpose of camera calibration in computer vision?
Correcting lens distortions
Which parameter is NOT typically estimated during camera calibrations?
Image Brightness
In Convolutional Neural Networks, some times we use “Transfer Learning”. What is Transfer Learning? In what situations, we use it? What advantages does transfer learning offer in CNNs?
Transfer learning is the process of using an existing model and retraining it to be used for a new purpose. Situations in which we use it are when there is a lack of training data,
