Machine Vision 1 TURN OVER Machine Vision, Alternative Assessment, replacing 2019/2020 Final Exam For: COMP0137 and COMPGI14 (for all current and previous students) You must answer ALL questions on ALL pages of this assignment. Marks for each part of each question are indicated in square brackets [n]. Implementing your own code and using slides, videos, and your own notes from this class ARE permitted. You may also access books, published papers, and existing Wikipedia articles online. But other online-resources, communications with others, and any collaboration are strictly NOT permitted. Posting or sharing your work with others, complete or incomplete, is NOT permitted. Further, you are expected to abide by the UCL Academic Manual, especially with respect to academic integrity. Section A: A Graphical Model 1. There is a Markov Model pictured below. Each discrete variable is illustrated to show what values are possible. A “configuration” is a specific combination, where each xi is in a specific state, for i=1..5. The unary costs, pairwise costs, and triplet costs for each variable or combination of variables are specified in the tables provided. All variables are discreet, but they do not all have the same number of possible states. For example, variable X1 can be “A” or “B”, while X2 can be “C”, “D”, or “E”. The unary costs for all five variables are: x 1 x 2 x 3 x 4 x 5 Machine Vision 2 CONTINUED The pairwise and triplet costs are here: Use the above information to answer the following questions. (a) Configuration A is ACGJN. Compute the total cost of this configuration: _____ [7 marks] (b) Configuration B is AEFKM. Compute the total cost of this configuration: _____ [7 marks] (c) What is the Maximum Likelihood configuration _ _ _ _ _ [7 marks] (d) What is the Maximum a Posteriori configuration _ _ _ _ _ [8 marks] [Total for Question 1: 29 marks] Section B: Cloud Tracking Task 2. You are installing two outdoor cameras to look up at the sky, each with a wide field of view. Your aim is to track some of the more distinct points on passing clouds, so you can infer their velocity in meters/second. An example image is shown below. [Question cont. over page] Machine Vision 3 TURN OVER [Question 2 cont.] “Clouds” Photo by Chris.L.Dodds,CC BY-SA 2.0 (a) You must choose the locations where the two cameras will be placed. Relative to the first camera, the second one should be placed what distance away Multiple choice: 0.5 m / 50 m / 500 m / 5000 m Answer: _______ [6 marks] (b) You must choose an interest-point detector, to find the distinct points that will be tracked by your system. Which detector would you choose Multiple choice: FAST / SUSAN / Canny / SIFT / DoG Answer: _____________ [6 marks] (c) Do the cameras need to be genlocked (“Genlock” stands for generator locking) Answer: __True or False ___ [4 marks] (d) Describe how telemetry of a single airplane helps you with this overall task. Comment on the need or benefit of having multiple airplanes. (Your answer must be 4 sentences or shorter.) ______________________________________________________________________________________________ ______________________________________________________________________________________________ _____________________________________________________________________________________________ Machine Vision 4 CONTINUED [Question 2 cont.] ____________________________________________________________________________________________ [7 marks] (e) How will your system deal with the fact that each cloud may be at a different altitude (Your answer must be 2 sentences or shorter.) ____________________________________________________________________________________________ ____________________________________________________________________________________________ [6 marks] [Total for Question 2: 29 marks] Section C: Miscellaneous Questions 3. Describe a unique computer vision classification task, naming three or more categories that you’d want to distinguish. The task must have severely imbalanced classes, and you must explain what you would do to cope with that imbalance. (Your answer must be 4 sentences or shorter.) ____________________________________________________________________________________________ ____________________________________________________________________________________________ ____________________________________________________________________________________________ ____________________________________________________________________________________________ [12 marks] 4. Does class imbalance have a counterpart in regression tasks What process would you follow to check for it (Your answer must be 3 sentences or shorter.) ____________________________________________________________________________________________ ____________________________________________________________________________________________ ____________________________________________________________________________________________ [9 marks] [Questions continued on next page] Machine Vision 5 END OF PAPER 5. You want to train a model with many parameters. Then you learn from an oracle that one (1) of the training labels is wrong. Describe one efficient way that may identify which training example had the wrong label (Your answer must be 3 sentences or shorter.) ____________________________________________________________________________________________ ____________________________________________________________________________________________ ____________________________________________________________________________________________ [9 marks] 6. Generative and Discriminative Models have different strengths and weaknesses. For each of six sub-questions, answer “G” or “D” (if Generative vs Discriminative is better) or “B” for Both, or “N” if neither is clearly preferable over the other. 6.a: Your input vector x is missing data in some dimensions: __________ 6.b: You have no access to a prior: __________ 6.c: You have access to a prior: __________ 6.d: Your model has only few parameters: __________ 6.e: You want to draw samples from your model: __________ 6.f: Your world-state is high-dimensional: __________ [12 marks total] [Total for Question 3, 4, 5, and 6: 42 marks] [Total for all questions: 100 marks]
