Skip to contentSkip to footer

Computer Vision Researcher Interview Questions

Salaries
Interviews

66 computer vision researcher interview questions shared by candidates

First part of the interview was kind of rapid fire of questions on machine learning basics: 01. Given a list of ML algorithms, briefly explain which ones belong to supervised and which ones belong to unsupervised. 02. Definition of precision, recall, and f1-score 03. How to synthesize data using KNN algorithm 04. How to do classification using Decision Tree 05. How to do normalization during data pre-processing 06. What is overfitting, what could cause it, and what would be the suggested solutions 07. What are MSE, OLS, MAP, MLP? 08. Does DNN suffer from curse of dimensionality? Why or why not? There should be about 10-12 questions, the rest of which I have forgotten.
avatar

Computer Vision Research Intern

Interviewed at Huawei Technologies

3.4★
Jan 30, 2023

First part of the interview was kind of rapid fire of questions on machine learning basics: 01. Given a list of ML algorithms, briefly explain which ones belong to supervised and which ones belong to unsupervised. 02. Definition of precision, recall, and f1-score 03. How to synthesize data using KNN algorithm 04. How to do classification using Decision Tree 05. How to do normalization during data pre-processing 06. What is overfitting, what could cause it, and what would be the suggested solutions 07. What are MSE, OLS, MAP, MLP? 08. Does DNN suffer from curse of dimensionality? Why or why not? There should be about 10-12 questions, the rest of which I have forgotten.

3. Take home test - implement a method of Facial landmark localization from a paper called 'Supervised Descent Method and its Applications to Face Alignment' from 2013. Note - the instruction they provide implies it's more of a coding task, but they actually expect you to have a deep understanding of everything: * How the SIFT/HoG descriptors are computed and why? * Why augmentations are needed? * What's better to have and why (in terms of prediction accuracy) - one model to predict all landmark points or a different model for each point.
avatar

Computer Vision Researcher

Interviewed at Lightricks

4★
Jan 13, 2025

3. Take home test - implement a method of Facial landmark localization from a paper called 'Supervised Descent Method and its Applications to Face Alignment' from 2013. Note - the instruction they provide implies it's more of a coding task, but they actually expect you to have a deep understanding of everything: * How the SIFT/HoG descriptors are computed and why? * Why augmentations are needed? * What's better to have and why (in terms of prediction accuracy) - one model to predict all landmark points or a different model for each point.

The 2nd part of the interview session was about 1 coding challenge. So the scenario was that given a square NxN matrix where N is an arbitrary positive integer, the elements represent the vaccination status on an imaginary city. So they are visualized to be located in a squarish manner. There are 3 statuses, namely, 0 represents unvaccinated, 1 represents vaccinated, 2 represents infected. If a person with status 0 is adjacent to at least 1 person with status 2 in his/her 4 cardinal points (up, down, left, right), then he/she will get infected. Person with status 2 at diagonal direction won't be able to infect other person. If person with status 0 is surrounded by people with status 1 at his/her 4 cardinal points, he/she won't get infected. Given this scenario, the question is to determine how many times does propagation of infection should happen so that all unvaccinated people in the NxN matrix are infected. Here, 1 infection propagation means refers to people with status 0 getting infected with people with status 2 for being directly adjacent to them. Other people with status 0 who are adjacent to the newly infected people will only be infected in the next infection propagation.
avatar

Computer Vision Research Intern

Interviewed at Huawei Technologies

3.4★
Jan 30, 2023

The 2nd part of the interview session was about 1 coding challenge. So the scenario was that given a square NxN matrix where N is an arbitrary positive integer, the elements represent the vaccination status on an imaginary city. So they are visualized to be located in a squarish manner. There are 3 statuses, namely, 0 represents unvaccinated, 1 represents vaccinated, 2 represents infected. If a person with status 0 is adjacent to at least 1 person with status 2 in his/her 4 cardinal points (up, down, left, right), then he/she will get infected. Person with status 2 at diagonal direction won't be able to infect other person. If person with status 0 is surrounded by people with status 1 at his/her 4 cardinal points, he/she won't get infected. Given this scenario, the question is to determine how many times does propagation of infection should happen so that all unvaccinated people in the NxN matrix are infected. Here, 1 infection propagation means refers to people with status 0 getting infected with people with status 2 for being directly adjacent to them. Other people with status 0 who are adjacent to the newly infected people will only be infected in the next infection propagation.

Viewing 1 - 10 interview questions

Glassdoor has 66 interview questions and reports from Computer vision researcher interviews. Prepare for your interview. Get hired. Love your job.