Skip to content
tezvyn:

Top 30 Easy Computer Vision Interview Questions and Answers for Freshers

30 easy multiple-choice Computer Vision interview questions, the ones an interviewer opens with: definitions, everyday syntax, and the quick checks that you have really used it. They come from 30 bites in the Computer Vision library, the gentlest slice of the 135 Computer Vision interview questions in the library. Answer them here or read straight down. Every question carries the correct option, why it is correct, and a link to the bite it came from.

Image/video models, diffusion, OCR, multimodal

30 questions. Pick an answer, or open “Show the answer” to read it.

Answers are graded in your browser. Nothing is saved, and no XP or streak is earned here. The app keeps score.

  1. Question 1 of 30

    Why is HSV often preferred over RGB for segmenting a colored object across a scene with mixed sunlight and shadow?

    Show the answer

    Answer: b · Hue stays relatively stable under lighting changes, so color separates from brightness

    HSV isolates hue from value, so a color keeps its hue under varying illumination, easing thresholding. Both spaces cover the same color gamut, so the capacity claim is false.

    Read the full bite: RGB versus HSV color spaces

  2. Question 2 of 30

    A photography platform archives original images for future editing and serves compressed previews. Which strategy best preserves fidelity while optimizing delivery?

    Show the answer

    Answer: d · Store originals in a lossless format and serve previews as lossy JPEG to reduce bandwidth.

    Lossless storage preserves bit-exact originals for future editing and avoids generational degradation, while lossy JPEG cuts preview sizes for faster web delivery. Option C is tempting because PNG guarantees exact pixels, but serving previews losslessly wastes bandwidth without perceptible quality gains over a well-compressed lossy image.

    Read the full bite: What is the difference between lossy and lossless image compression?

  3. Question 3 of 30

    Which statement best describes how histogram equalization remaps grayscale intensities?

    Show the answer

    Answer: d · It applies the cumulative distribution function to redistribute intensities toward a uniform histogram.

    Histogram equalization uses the cumulative distribution function of the original histogram to remap intensities so the output approximates a uniform distribution, maximizing global contrast. The first option describes linear contrast stretching, which merely rescales the minimum and maximum values without considering the frequency of each intensity level.

    Read the full bite: Describe a grayscale histogram and its use in exposure and equalization

  4. Question 4 of 30

    When implementing a box blur, why is it important to write results into a separate destination buffer rather than updating the source image in place?

    Show the answer

    Answer: c · It prevents already-blurred pixel values from being reused in later neighborhood averages

    Using a separate destination buffer guarantees that every neighborhood average reads only original pixel values, not values that have already been blurred and would distort subsequent averages. The overflow issue in option B is addressed by using a larger type for the accumulator during the sum, not by allocating a second image buffer.

    Read the full bite: How would you implement a simple box blur on a grayscale image?

  5. Question 5 of 30

    A grayscale image looks dull because most pixels are clustered between intensity 100 and 150. After histogram equalization, what has fundamentally changed about the pixel intensities?

    Show the answer

    Answer: c · A transfer function based on the cumulative intensity distribution was used as a lookup table to spread values across the full range.

    Histogram equalization computes the cumulative distribution function from the histogram, normalizes it to the maximum intensity, and uses it as a lookup table to remap pixels across the full range. Option A describes linear contrast stretching, which only scales the min and max values without considering the actual probability distribution of intensities.

    Read the full bite: What is an image histogram and how does histogram equalization improve contrast?

  6. Question 6 of 30

    Why does the Harris detector consider a corner a more reliable tracking feature than a point along a straight edge?

    Show the answer

    Answer: a · A corner constrains position in two directions, while an edge point can slide along the edge

    At a corner, intensity changes in every shift direction, giving large eigenvalues and a precise 2D location. Along an edge only one direction constrains position, so the point slides (the aperture problem).

    Read the full bite: Harris corner detector and corner stability

  7. Question 7 of 30

    How does the Canny edge detector relate to the Sobel operator in a typical pipeline?

    Show the answer

    Answer: a · Canny uses Sobel-style gradients, then adds non-maximum suppression and hysteresis thresholding

    Canny is a multi-stage pipeline that computes gradients (as Sobel does), thins them with non-maximum suppression, and links them with double-threshold hysteresis. Sobel alone gives thick, noisy edges.

    Read the full bite: Image gradients, Sobel, and Canny

  8. Question 8 of 30

    Knowing point p1 in the first image, why does the epipolar constraint reduce the search for its match p2 to a single line in the second image?

    Show the answer

    Answer: d · Because p1's 3D point lies along one viewing ray, which projects to a line in the second image

    The unknown depth of p1 means its 3D point spans a ray, and that ray projects to the epipolar line in the second image where p2 must lie. Identical intrinsics or rectification are not required for the constraint to hold.

    Read the full bite: Epipolar constraint for correspondence search

  9. Question 9 of 30

    In a rectified stereo pair, how does the depth of a scene point relate to its disparity?

    Show the answer

    Answer: a · Depth is inversely proportional to disparity, given the baseline and focal length

    Depth equals focal length times baseline divided by disparity, so larger disparity means a nearer point. The direct-proportion and independence options invert or ignore this geometric relationship.

    Read the full bite: Disparity and depth in stereo vision

  10. Question 10 of 30

    Which property of a convolutional layer most directly explains why it has far fewer parameters than a dense layer over the same input?

    Show the answer

    Answer: c · The same kernel weights are reused at every spatial position

    Weight sharing means one small kernel is applied across all positions, so parameters scale with kernel size not input size. Padding and stride affect output dimensions, not parameter count, and full connectivity is exactly what conv layers avoid.

    Read the full bite: How a convolutional layer works

  11. Question 11 of 30

    What is a key advantage of a strided convolution over max pooling for downsampling?

    Show the answer

    Answer: d · It learns how to downsample instead of fixing a non-learnable rule

    Strided convolution learns its downsampling jointly with feature extraction, unlike the fixed max rule. It actually adds parameters rather than saving them, and neither method guarantees full translation invariance.

    Read the full bite: Max pooling versus strided convolution

  12. Question 12 of 30

    If every activation function were removed from a deep CNN, what would happen to its representational power?

    Show the answer

    Answer: d · It would collapse to the power of a single linear layer

    A composition of purely linear operations is itself one linear map, so any depth reduces to a single linear layer that can only model linear boundaries. Nonlinearity, not depth alone, is what grants expressive power.

    Read the full bite: Why CNNs need nonlinear activations like ReLU

  13. Question 13 of 30

    What does object detection provide that image classification does not?

    Show the answer

    Answer: a · Localization of multiple objects via bounding boxes with class labels

    Detection adds localization and handles multiple objects, outputting boxes with labels and scores. A single whole-image label is classification, and pixel-level masks describe segmentation rather than detection.

    Read the full bite: Image classification versus object detection

  14. Question 14 of 30

    How is Intersection over Union computed for a predicted and ground-truth bounding box?

    Show the answer

    Answer: c · Intersection area divided by the union of both boxes' areas

    IoU divides the overlapping area by the combined area of both boxes, ranging zero to one. Dividing by only the predicted area, inverting the ratio, or using center distance all describe different, incorrect metrics.

    Read the full bite: Intersection over Union for detection

  15. Question 15 of 30

    What makes COCO's primary mAP metric stricter than a single-threshold detection score?

    Show the answer

    Answer: a · It averages Average Precision over multiple IoU thresholds, rewarding tight localization

    COCO averages AP across IoU thresholds from 0.5 to 0.95, so loose boxes are penalized. It does not threshold on confidence at 0.95, drop rare classes, or use pixel masks for the box-based metric.

    Read the full bite: Mean Average Precision in object detection

  16. Question 16 of 30

    What can instance segmentation do that semantic segmentation cannot?

    Show the answer

    Answer: c · Separate individual objects of the same class into distinct masks

    Instance segmentation distinguishes separate objects of the same class, which semantic segmentation merges into one labeled region. Per-pixel class labeling is what both share, and boxes or whole-image labels describe detection and classification.

    Read the full bite: Semantic versus instance segmentation

  17. Question 17 of 30

    Why does the Lucas-Kanade method assume all pixels in a small window share the same motion?

    Show the answer

    Answer: a · To resolve the aperture problem by making the system overdetermined

    A single pixel gives one equation in two unknowns (the aperture problem); assuming a window of pixels moves identically yields an overdetermined system solvable by least squares. It does not concern grayscale conversion or removing gradients.

    Read the full bite: Sparse vs dense optical flow and Lucas-Kanade

  18. Question 18 of 30

    What core problem must a tracking-by-detection system solve that a per-frame detector does not?

    Show the answer

    Answer: c · Associating detections across frames to maintain consistent identities

    Detectors output independent boxes with no identity, so the tracker must link detections across frames into consistent tracks via data association. Mask conversion and camera calibration are unrelated to maintaining identity over time.

    Read the full bite: Design a tracking-by-detection tracker

  19. Question 19 of 30

    What is the first transformation a ViT applies to a raw image that a CNN does not?

    Show the answer

    Answer: b · Splitting the image into patches and linearly embedding each as a token

    A ViT partitions the image into fixed patches and projects each into a token embedding, whereas a CNN's first step is sliding convolutional filters. Pixel normalization is a generic preprocessing step common to both, not the distinguishing operation.

    Read the full bite: How ViT and CNN process an image differently

  20. Question 20 of 30

    How is a single patch's self-attention output computed from queries, keys, and values?

    Show the answer

    Answer: a · Score the patch's query against all keys, softmax the scores, and weight-sum all values

    The patch's query is dot-producted with every key to get relevance scores, which are scaled and softmaxed into weights used to combine all value vectors. Equal averaging or max-key copying would discard the learned, content-dependent weighting that defines attention.

    Read the full bite: Self-attention over image patches explained

  21. Question 21 of 30

    During GAN training, what is the generator directly optimizing for?

    Show the answer

    Answer: a · Maximizing the probability the discriminator labels its fakes as real

    The generator never sees real images directly; it learns by trying to fool the discriminator into calling its outputs real. Reconstruction error is an autoencoder objective, and classifying real versus fake is the discriminator's job.

    Read the full bite: GAN architecture: generator and discriminator roles

  22. Question 22 of 30

    In a standard diffusion model, which component is learned by the neural network?

    Show the answer

    Answer: d · The reverse denoising process

    The forward process is a fixed, non-learned noise schedule, while the reverse denoising process is what the network learns. Claiming the forward process is learned is the most common misconception.

    Read the full bite: Diffusion forward and reverse processes

  23. Question 23 of 30

    Which segmentation type assigns every pixel a class while also distinguishing individual object instances?

    Show the answer

    Answer: d · Panoptic segmentation

    Panoptic segmentation unifies stuff labeling with per-instance thing identity over all pixels. Semantic segmentation labels classes but not instances, and classic instance segmentation often ignores background stuff.

    Read the full bite: Semantic, instance, and panoptic segmentation

  24. Question 24 of 30

    In a simple VQA baseline, how is producing the answer typically framed?

    Show the answer

    Answer: c · Classification over a fixed vocabulary of frequent answers

    Baseline VQA treats answering as multi-class classification over the most common answers, which is simple and trainable with cross-entropy. Free-form generation is a more complex later approach, not the simple baseline.

    Read the full bite: Designing a baseline Visual Question Answering model

  25. Question 25 of 30

    With only a small labeled dataset, what is the safest way to adapt a pretrained ResNet50?

    Show the answer

    Answer: d · Freeze the backbone, replace the head, train it, then optionally fine-tune top blocks at a low rate

    Freezing the generic backbone and training a new head avoids overfitting on small data, with cautious top-layer fine-tuning as an optional gain. Retraining everything overfits, and the early layers hold the most transferable features, so they should stay frozen.

    Read the full bite: Transfer learning from ResNet50 on small data

  26. Question 26 of 30

    What is the most effective high-level way to use 1 million unlabeled images with only 1,000 labels?

    Show the answer

    Answer: b · Self-supervised pretraining on all images, then fine-tune on the labels

    Self-supervised pretraining learns strong representations from the unlabeled set so the few labels suffice to fit a classifier. Discarding the unlabeled data wastes the main asset and overfits on a thousand examples.

    Read the full bite: Leveraging unlabeled data with limited labels

  27. Question 27 of 30

    Why must the Camera Response Function be recovered before merging bracketed exposures into an HDR radiance map?

    Show the answer

    Answer: a · Pixel values are a nonlinear function of radiance and must be linearized

    The CRF is the nonlinear map from radiance to pixel value; inverting it linearizes the data so exposures can be merged correctly. Registration and range compression (tone mapping) are separate steps.

    Read the full bite: How do you build an HDR image from bracketed exposures?

  28. Question 28 of 30

    Why is RANSAC used when estimating the homography between two overlapping panorama images?

    Show the answer

    Answer: d · It robustly fits the transform despite many incorrect feature matches

    Feature matching produces many outliers; RANSAC fits the homography from random minimal samples and keeps the model with the most inliers. Blending and projection are unrelated later stages.

    Read the full bite: Outline the classic image stitching pipeline.

  29. Question 29 of 30

    Why should data augmentation be applied to the training set but not the validation or test set?

    Show the answer

    Answer: c · Evaluation must reflect real, unaltered data to measure true performance

    Augmentation regularizes training, but validation and test sets must mirror real inputs so metrics are honest. Speed and dataset size are not the reason, and validation data is labeled.

    Read the full bite: What data augmentations help small image datasets?

  30. Question 30 of 30

    For a self-driving car's pedestrian detector, which metric should usually be prioritized and why?

    Show the answer

    Answer: a · Recall, because failing to detect a real pedestrian is dangerous

    Missing a pedestrian can be fatal, so high recall matters even at the cost of some false alarms. Prioritizing precision would risk dangerous misses, and the two genuinely trade off.

    Read the full bite: Precision vs recall in object detection.

Could you explain these out loud?

That is what an interview actually tests. Tezvyn gives you questions like these with what the interviewer is really checking, the answer that lands, and the mistake that ends the conversation, in the four minutes before your next meeting.

The iPhone app is on the way

We are building it. Until it lands, nothing here is held back from you: every interview card, your saved cards, streaks and the job board all work in Safari, plus hundreds of free practice quizzes of thirty questions each. Sign in and it all carries over to the app the day it arrives.

Want it as an icon? Tap Share at the bottom of Safari, then Add to Home Screen. It opens full screen and the cards you have read stay available offline.

Get it on Google PlayiPhone app coming soon