Top 30 Advanced Computer Vision Concepts Quiz

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

   Which statement best explains why radiometry is crucial for a self-driving car's camera system?

   • aIt simplifies the light measurement process by ignoring complex spectral responses.
   • bIt provides data that aligns with how human drivers perceive the road, improving safety.
   • cIt uses lumens, which are the standard units for all digital imaging sensors.
   • dIt quantifies the physical energy of light hitting the sensor, enabling objective machine interpretation.
   Show the answer

   Answer: d · It quantifies the physical energy of light hitting the sensor, enabling objective machine interpretation.

   Radiometry measures the actual physical energy of light (in watts), which is critical for machines like self-driving cars to objectively interpret their environment, independent of human perception. Option B is incorrect because radiometry specifically avoids human perception weighting, which is the domain of photometry.

   Read the full bite: Radiometry: Measuring Light as Physics, Not Perception

2. Question 2 of 30

   Which optical phenomenon is NOT directly accounted for by a standard Bidirectional Reflectance Distribution Function (BRDF)?

   • aThe way light penetrates and exits a marble sculpture.
   • bThe uniform scattering of light from a matte wall.
   • cThe directional sheen observed on brushed metal.
   • dThe mirror-like reflection from a polished surface.
   Show the answer

   Answer: a · The way light penetrates and exits a marble sculpture.

   The card explicitly states that standard BRDF models are for opaque surfaces only and do not account for light transmitted through a material (translucency) or scattered underneath the surface, which describes light interacting with a marble sculpture. The other options are all forms of surface reflection that BRDFs are designed to model.

   Read the full bite: BRDF: Modeling How Surfaces Reflect Light

3. Question 3 of 30

   What is the fundamental consequence of a digital camera utilizing a Bayer filter for its image sensor?

   • aIt requires three separate image sensors, each dedicated to capturing a single primary color.
   • bThe sensor captures a monochrome image, which is later colorized by software.
   • cEach photosite records only one color component, requiring software interpolation for full-color pixels.
   • dEvery photosite on the sensor directly captures full RGB color information.
   Show the answer

   Answer: c · Each photosite records only one color component, requiring software interpolation for full-color pixels.

   The card states that a Bayer filter allows each pixel to see only one color, and the full-color image is then reconstructed in software through a process called demosaicing. Option D is incorrect because the card explicitly states a Bayer filter does not provide 24 million full-color pixels directly.

   Read the full bite: Bayer Filter: Faking Full Color with a Filter Grid

4. Question 4 of 30

   What is the primary advantage of the Canny edge detector over simpler methods like the Sobel filter?

   • aIt is significantly faster due to its optimized gradient calculation.
   • bIt produces clean, thin, and continuous edge maps by effectively managing noise and connecting weak edges.
   • cIt completely eliminates all noise from an image before edge detection.
   • dIt requires no parameter tuning, making it easier to apply universally.
   Show the answer

   Answer: b · It produces clean, thin, and continuous edge maps by effectively managing noise and connecting weak edges.

   The card highlights that Canny's design provides a more robust method that finds continuous, well-defined edges, which is achieved through its multi-stage process of noise reduction, thinning (non-maximum suppression), and connecting weak edges (hysteresis). Option A is incorrect because the card states simpler filters might be preferred if raw speed is the only priority, implying Canny is not faster.

   Read the full bite: Canny Edge Detector: Finding True Edges

5. Question 5 of 30

   Which visual effect cannot be achieved using only an affine transformation?

   • aSkewing a square into a parallelogram
   • bRotating an image around its center point
   • cResizing an image while maintaining its aspect ratio
   • dMaking an object appear to shrink as it moves further away
   Show the answer

   Answer: d · Making an object appear to shrink as it moves further away

   Affine transformations preserve parallelism, meaning parallel lines remain parallel, which prevents them from modeling perspective effects like objects appearing smaller with distance. This requires a more general projective transformation. Rotating an image around its center point, while involving both rotation and translation, is fully supported by affine transformations.

   Read the full bite: Affine Transformations: Move, Scale, and Rotate Without Bending

6. Question 6 of 30

   How does the Laplacian of Gaussian (LoG) primarily identify the center of a blob in an image?

   • aBy finding local extrema (peaks or valleys) after applying a Gaussian blur and then the Laplacian operator.
   • bBy detecting zero-crossings in the output of the combined Gaussian and Laplacian operations.
   • cBy directly convolving the image with a Difference of Gaussians (DoG) kernel.
   • dBy identifying areas where the image intensity gradient is steepest.
   Show the answer

   Answer: a · By finding local extrema (peaks or valleys) after applying a Gaussian blur and then the Laplacian operator.

   The card states that after applying the Gaussian blur and Laplacian, 'the local extrema (peaks and valleys) mark their centers.' Option B is incorrect because zero-crossings indicate the edges of blobs, not their centers.

   Read the full bite: Laplacian of Gaussian (LoG) for Blob Detection

7. Question 7 of 30

   What is the primary non-technical factor that might prevent a commercial project from utilizing SURF?

   • aIt lacks the ability to reliably identify features under diverse lighting conditions.
   • bIts computational demands often exceed the capabilities of embedded systems.
   • cThe underlying algorithms are patented, necessitating licensing for commercial deployment.
   • dIt offers reduced robustness to image scaling and rotation compared to SIFT.
   Show the answer

   Answer: c · The underlying algorithms are patented, necessitating licensing for commercial deployment.

   The card explicitly states that the primary reason to avoid SURF in commercial products is its patented algorithms, requiring licenses. The other options describe technical limitations that SURF is designed to overcome or perform well against, making them incorrect.

   Read the full bite: SURF: A Faster, Patented Alternative to SIFT

8. Question 8 of 30

   What is ORB's primary mechanism for making its feature descriptors robust to in-plane rotation?

   • aIt employs a descriptor that is fundamentally different from BRIEF, designed for rotation invariance from scratch.
   • bIt builds an image pyramid, allowing it to detect features at all possible rotational angles.
   • cIt computes an orientation for each keypoint and rotates the BRIEF sampling pattern accordingly.
   • dIt uses the FAST algorithm, which inherently extracts rotation-invariant keypoints.
   Show the answer

   Answer: c · It computes an orientation for each keypoint and rotates the BRIEF sampling pattern accordingly.

   ORB achieves rotation invariance by first computing an orientation for each keypoint and then rotating, or 'steering,' the BRIEF descriptor's sampling pattern to match that orientation. Option D is incorrect because FAST detects keypoints, but the rotation invariance for the descriptor is added in the subsequent BRIEF modification.

   Read the full bite: ORB: Fast, Free Feature Detection for Computer Vision

9. Question 9 of 30

   According to the card, what is the primary reason for combining Local Binary Patterns (LBP) with other descriptors like HOG?

   • aTo allow LBP to process images that contain only global structural information, rather than local textures.
   • bTo achieve significantly improved performance in computer vision tasks, such as object detection.
   • cTo overcome LBP's inherent limitation in extracting any meaningful texture features from an image.
   • dTo decrease the total number of features generated, thereby speeding up the machine learning model's training process.
   Show the answer

   Answer: b · To achieve significantly improved performance in computer vision tasks, such as object detection.

   The card explicitly states that combining LBP with other descriptors like HOG is done "For even better results, especially in object detection." This directly supports option B. Option C is incorrect because LBP is designed to provide powerful texture descriptions; combining it enhances performance, it doesn't compensate for an inability to extract features.

   Read the full bite: Local Binary Patterns (LBP) for Texture Classification

10. Question 10 of 30

    An engineer switches from BFMatcher to FlannBasedMatcher in an OpenCV ORB pipeline and preserves the normType and crossCheck arguments. What is the most likely outcome?

    • aFLANN silently ignores the unsupported Brute-Force parameters and uses its own defaults, producing unexpected matches without warning.
    • bFLANN automatically converts the Hamming norm and crossCheck flag into equivalent internal index parameters.
    • cOpenCV raises a runtime exception because FLANN does not accept normType or crossCheck.
    • dThe matcher falls back to Brute-Force behavior because ORB descriptors require Hamming distance.
    Show the answer

    Answer: a · FLANN silently ignores the unsupported Brute-Force parameters and uses its own defaults, producing unexpected matches without warning.

    The card warns that engineers often assume FLANN shares Brute-Force's normType and crossCheck parameters, causing silent configuration errors rather than exceptions or automatic adaptation. Option C is tempting because invalid arguments usually raise errors, but the source emphasizes that this mismatch fails silently.

    Read the full bite: FLANN Matcher for Feature Correspondence

11. Question 11 of 30

    What is the primary reason the original Eight-Point Algorithm is not recommended for estimating the fundamental matrix in practical scenarios?

    • aIt is numerically unstable, often producing inaccurate results for the fundamental matrix.
    • bIt inherently assumes calibrated cameras, which contradicts the purpose of the fundamental matrix.
    • cIt is computationally more expensive than the normalized eight-point algorithm by Hartley.
    • dIt requires a minimum of ten corresponding points for robust fundamental matrix estimation.
    Show the answer

    Answer: a · It is numerically unstable, often producing inaccurate results for the fundamental matrix.

    The card explicitly states that the original algorithm is numerically unstable when used to compute the fundamental matrix, leading to its avoidance in practical settings. While the essential matrix (which the original algorithm estimates well) requires calibrated cameras, the instability is the specific reason given for not using the original algorithm for the fundamental matrix.

    Read the full bite: Eight-Point Algorithm: Finding Geometry from Image Pairs

12. Question 12 of 30

    What is the primary distinguishing characteristic of Bundle Adjustment in a 3D reconstruction pipeline?

    • aIts ability to rapidly generate initial estimates of camera poses and 3D points for real-time systems.
    • bIt focuses exclusively on refining the intrinsic parameters of a single camera to correct lens distortions.
    • cIt primarily triangulates 3D points from two-view correspondences, providing a robust initial 3D structure.
    • dIt performs a simultaneous, global optimization of 3D point locations, camera poses, and intrinsic parameters to minimize reprojection error.
    Show the answer

    Answer: d · It performs a simultaneous, global optimization of 3D point locations, camera poses, and intrinsic parameters to minimize reprojection error.

    Bundle Adjustment's core function is to jointly optimize all elements—3D points, camera poses, and intrinsic parameters—to collectively minimize the reprojection error. It is explicitly stated as computationally expensive and not suitable for real-time initial estimation, making option A incorrect.

    Read the full bite: Bundle Adjustment: Jointly Refining 3D Scenes and Cameras

13. Question 13 of 30

    What fundamental challenge does SLAM overcome that simpler navigation methods cannot?

    • aAccurately tracking an agent's position using only internal motion sensors.
    • bCreating a detailed, persistent map of a known environment for future use.
    • cSimultaneously building a map of an unknown space while determining the agent's location within that evolving map.
    • dIntegrating data from multiple sensor types to improve positional accuracy.
    Show the answer

    Answer: c · Simultaneously building a map of an unknown space while determining the agent's location within that evolving map.

    SLAM's core innovation is its ability to concurrently construct a map of an unknown environment and localize an agent within that developing map. This addresses the 'chicken-and-egg' problem that simpler methods fail to solve.

    Read the full bite: SLAM: Mapping a Room While You're Still In It

14. Question 14 of 30

    What is the primary purpose of the learned scaling factor (gamma) and shifting factor (beta) in Batch Normalization?

    • aTo ensure that the batch statistics are reliably estimated even with very small batch sizes.
    • bTo re-center the activations to a mean of zero and re-scale them to a variance of one.
    • cTo enable the network to learn and apply an optimal mean and variance for each layer's inputs.
    • dTo dynamically adjust the learning rate for each neuron based on its activation statistics.
    Show the answer

    Answer: c · To enable the network to learn and apply an optimal mean and variance for each layer's inputs.

    After the initial normalization to mean zero and variance one, gamma and beta allow the network to learn and apply an optimal scale and mean for the activations, potentially deviating from zero and one if it aids training. Option B describes the initial normalization step, which occurs before gamma and beta are applied.

    Read the full bite: Batch Normalization: Stabilizing Neural Network Training

15. Question 15 of 30

    What is the primary mechanism by which Residual Networks (ResNets) overcome the accuracy degradation problem in very deep neural networks?

    • aThey prevent overfitting by reducing the total number of trainable parameters in deep architectures.
    • bThey use smaller kernel sizes in convolutional layers to extract finer-grained features.
    • cThey enable layers to learn a residual function F(x) that adjusts the input x, rather than learning a complete mapping H(x).
    • dThey introduce additional non-linear activation functions to increase model complexity.
    Show the answer

    Answer: c · They enable layers to learn a residual function F(x) that adjusts the input x, rather than learning a complete mapping H(x).

    ResNets address the degradation problem by having layers learn a small correction (residual) F(x) to the input x, rather than the entire output H(x). This makes it easier for the network to learn the identity function or small adjustments. Distractor C is incorrect because the card explicitly states that degradation was not overfitting, and ResNets enable deeper networks, which typically means more parameters.

    Read the full bite: Residual Networks (ResNets): Go Deeper, Not Worse

16. Question 16 of 30

    What is the primary function of the 1x1 convolution used as a "bottleneck" before the 3x3 and 5x5 convolutions in an Inception module?

    • aTo increase the number of feature maps, enriching the representation.
    • bTo introduce non-linear activation functions, enhancing model capacity.
    • cTo expand the receptive field of the module without increasing parameters significantly.
    • dTo reduce the channel depth of the input, making subsequent operations more computationally efficient.
    Show the answer

    Answer: d · To reduce the channel depth of the input, making subsequent operations more computationally efficient.

    The 1x1 bottleneck convolution's main purpose is to reduce the input's channel depth, which drastically cuts down the computational expense of the following larger convolutions. While 1x1 convolutions can capture local features, their specific role as a bottleneck is not to expand the receptive field, but to manage computational load.

    Read the full bite: GoogLeNet's Inception Module: Wider, Not Just Deeper

17. Question 17 of 30

    What is the primary advantage of using transfer learning for computer vision tasks when your specific dataset is small?

    • aIt simplifies the neural network architecture, making it easier to train on limited resources.
    • bIt automatically augments the small dataset with synthetic images, increasing its size.
    • cIt allows the model to be trained from scratch on a smaller dataset, which is faster.
    • dIt leverages general features learned from a large, diverse dataset, reducing the need for extensive specific data.
    Show the answer

    Answer: d · It leverages general features learned from a large, diverse dataset, reducing the need for extensive specific data.

    Transfer learning's main advantage is reusing foundational knowledge (general features like edges and textures) learned from a massive dataset, meaning the model needs less new data to adapt to specific patterns. Option C is incorrect because transfer learning explicitly avoids training from scratch.

    Read the full bite: Transfer Learning: Don't Train Vision Models from Scratch

18. Question 18 of 30

    Why does SSD suffer degraded small-object accuracy when predictions from shallow feature maps are omitted?

    • aEarly layers encode high-level semantic features that deeper convolutional maps lose through downsampling.
    • bShallow feature maps primarily reduce computational cost by pre-filtering default boxes before deep layers process them.
    • cEarly layers retain fine spatial resolution necessary for detecting small objects, so omitting them degrades localization accuracy.
    • dShallow layers generate region proposals that deeper layers refine during the single forward pass.
    Show the answer

    Answer: c · Early layers retain fine spatial resolution necessary for detecting small objects, so omitting them degrades localization accuracy.

    The card explains that early layers carry fine detail and that combining multiple feature map resolutions handles size variation, so discarding shallow maps removes high-resolution predictions needed for small objects. Option D is tempting because detection systems often use proposals, but SSD explicitly eliminates all proposal generation to remain a single network.

    Read the full bite: SSD: Real-Time Detection Without Region Proposals

19. Question 19 of 30

    Which statement best describes how Focal Loss addresses class imbalance during model training?

    • aIt dynamically scales down the loss contribution from easily classified examples.
    • bIt assigns a fixed, higher weight to examples from underrepresented classes.
    • cIt introduces a separate regularization term to prevent overfitting to background samples.
    • dIt oversamples minority class examples to increase their representation.
    Show the answer

    Answer: a · It dynamically scales down the loss contribution from easily classified examples.

    Focal Loss works by dynamically reducing the loss for well-classified (easy) examples, allowing the model to focus its training capacity on hard examples. While oversampling or fixed weighting (options A and B) are common strategies for class imbalance, they are not the mechanism of Focal Loss, which modifies the loss function itself based on classification difficulty.

    Read the full bite: Focal Loss: Forcing Models to Learn from Hard Examples

20. Question 20 of 30

    In which scenario would instance segmentation be the most appropriate computer vision technique?

    • aDrawing bounding boxes around every car and pedestrian in a busy street scene.
    • bPrecisely outlining and counting each individual tumor cell in a medical image for analysis.
    • cIdentifying all pixels belonging to the "road" or "sky" in an autonomous vehicle's camera feed.
    • dDetermining if any "person" is present in a security camera's field of view.
    Show the answer

    Answer: b · Precisely outlining and counting each individual tumor cell in a medical image for analysis.

    Instance segmentation is ideal for tasks requiring the precise outlining and counting of individual objects, such as distinct tumor cells. Option C describes semantic segmentation, while options B and D are better suited for object detection, which does not provide pixel-level masks for individual instances.

    Read the full bite: Instance Segmentation: Counting and Outlining Objects

21. Question 21 of 30

    Which application most clearly demonstrates the unique advantage of panoptic segmentation over its predecessors?

    • aCategorizing every pixel in an image as either 'sky', 'water', or 'land'.
    • bCounting the exact number of individual 'pedestrians' in a crowded scene.
    • cA robot needing to identify 'cup #1' on a 'table' while avoiding 'cup #2' and recognizing the 'wall' behind it.
    • dMapping all pixels that represent a 'road' surface for navigation.
    Show the answer

    Answer: c · A robot needing to identify 'cup #1' on a 'table' while avoiding 'cup #2' and recognizing the 'wall' behind it.

    Panoptic segmentation unifies semantic and instance segmentation, providing both class labels for 'stuff' (like 'table' or 'wall') and unique instance IDs for 'things' (like 'cup #1' and 'cup #2'). Option C requires this combined understanding. Options A and C describe semantic segmentation, while Option B describes instance segmentation.

    Read the full bite: Panoptic Segmentation: A Unified View of a Scene

22. Question 22 of 30

    What is the primary advancement DeepLab offers over traditional object detection methods?

    • aIt can identify objects at multiple scales simultaneously.
    • bIt accurately categorizes the overall scene depicted in an image.
    • cIt assigns a precise class label to every individual pixel in an image.
    • dIt significantly reduces the computational cost for image processing.
    Show the answer

    Answer: c · It assigns a precise class label to every individual pixel in an image.

    DeepLab's fundamental contribution is pixel-level semantic segmentation, meaning it classifies every pixel, creating a precise mask for objects, unlike object detection which only provides bounding boxes. While DeepLab does incorporate techniques for multi-scale understanding and efficiency, its defining advancement over object detection is this granular, pixel-accurate output.

    Read the full bite: DeepLab: Pixel-Level Semantic Image Segmentation

23. Question 23 of 30

    In the R-CNN family, why is performing selective search on CNN feature maps rather than raw pixels considered a critical design win?

    • aIt guarantees that each proposed region will contain exactly one object instance, eliminating duplicate detections.
    • bThe same deep convolutional representations are reused for localizing objects and classifying them, rather than computing features twice.
    • cRegion proposals inherit hierarchical visual patterns from the feature maps, allowing category assignment during the proposal stage itself.
    • dIt allows bounding box coordinates to be predicted directly by the CNN without requiring a separate region proposal stage.
    Show the answer

    Answer: b · The same deep convolutional representations are reused for localizing objects and classifying them, rather than computing features twice.

    The card emphasizes that operating selective search on convolutional features enables the model to reuse the same deep representations for both localization and classification. Distractor B is tempting because the card mentions hierarchical patterns, but it incorrectly conflates the stages; the card explicitly states selective search draws boxes without yet knowing their exact categories.

    Read the full bite: Mask R-CNN: Region-Based Detection

24. Question 24 of 30

    How does a Conditional Random Field (CRF) primarily achieve local consistency in structured prediction tasks like image segmentation?

    • aBy training a separate classifier for each distinct region in the input data.
    • bBy performing multiple independent classifications and selecting the most frequent outcome.
    • cBy allowing each data point to vote on its own label based on its unique features.
    • dBy penalizing neighboring elements that are assigned different labels.
    Show the answer

    Answer: d · By penalizing neighboring elements that are assigned different labels.

    The card explicitly states that a CRF's score includes a 'pairwise term' which 'penalizes adjacent pixels for having different labels,' directly enforcing local consistency. While individual features (unary term) are considered, the pairwise term is the mechanism for ensuring consistency across neighbors.

    Read the full bite: Conditional Random Fields: Labeling with Context

25. Question 25 of 30

    What is the fundamental advantage of a Two-Stream ConvNet over a single-stream ConvNet for video action recognition?

    • aIt eliminates the need for pre-processing steps like optical flow calculation.
    • bIt allows for the explicit and separate modeling of static visual content and dynamic motion patterns.
    • cIt can process videos of arbitrary length without requiring fixed-size input.
    • dIt significantly reduces the computational overhead by processing frames in parallel.
    Show the answer

    Answer: b · It allows for the explicit and separate modeling of static visual content and dynamic motion patterns.

    The core advantage is its ability to explicitly model two distinct information sources: object appearance (spatial stream) and motion (temporal stream via optical flow). Option D is incorrect because the architecture has a high computational cost due to optical flow pre-computation.

    Read the full bite: Two-Stream ConvNets: Seeing What vs. How It Moves

26. Question 26 of 30

    Which mechanism is central to how Correlation Filter-based Trackers (CFTs) efficiently locate an object in subsequent video frames?

    • aRe-detecting the object using a pre-trained classifier in each frame.
    • bSegmenting the image into regions and classifying the object's region based on its appearance.
    • cPredicting the object's movement based on its past trajectory and velocity.
    • dIdentifying the peak response in a map generated from applying a learned filter to a search area.
    Show the answer

    Answer: d · Identifying the peak response in a map generated from applying a learned filter to a search area.

    CFTs operate by applying a learned filter (template) across a search area in a new frame and identifying the location of the maximum value in the resulting response map, which corresponds to the object's predicted position. Re-detecting the object with a classifier in every frame would be computationally intensive, which CFTs aim to avoid for speed.

    Read the full bite: Correlation Filters: Fast and Robust Object Tracking

27. Question 27 of 30

    A particle filter is particularly effective for object tracking in scenarios characterized by significant ambiguity or unpredictable movement because it:

    • aReduces computational overhead compared to single-hypothesis trackers.
    • bAssumes a perfectly linear motion model, simplifying predictions.
    • cOnly requires initial object detection, eliminating the need for continuous sensor updates.
    • dSimultaneously evaluates multiple potential object states, enabling robust handling of multi-modal uncertainty.
    Show the answer

    Answer: d · Simultaneously evaluates multiple potential object states, enabling robust handling of multi-modal uncertainty.

    The card states that particle filters maintain a 'cloud of possible states' and handle 'multi-modal uncertainty,' which allows them to track objects through clutter and occlusion where a single best guess would fail. Option A is incorrect because particle filters can be computationally intensive due to the 'curse of dimensionality'.

    Read the full bite: Particle Filters for Object Tracking

28. Question 28 of 30

    DeepSORT significantly improves object tracking robustness, especially during occlusions. What is its key mechanism for maintaining consistent object identities in such scenarios?

    • aBy exclusively using a highly optimized Kalman filter to predict object trajectories.
    • bBy increasing the sensitivity of the object detector to find partially occluded objects.
    • cBy generating and comparing unique appearance descriptors using a deep neural network.
    • dBy assigning a new, temporary ID to objects that re-emerge and merging tracks later.
    Show the answer

    Answer: c · By generating and comparing unique appearance descriptors using a deep neural network.

    The card states DeepSORT calculates an 'appearance descriptor' using a pre-trained convolutional neural network to match objects based on appearance features, which allows it to 'bridge long occlusions'. Option A is incorrect because while DeepSORT uses a Kalman filter, the card explains that simple motion prediction 'fails' during long occlusions, necessitating the appearance metric.

    Read the full bite: DeepSORT: Adding Visual Memory to Object Tracking

29. Question 29 of 30

    Which approach allows Temporal Segment Networks (TSN) to effectively capture long-range temporal context for action recognition?

    • aGenerating motion vectors between adjacent frames to represent dynamic changes over time.
    • bEmploying 3D convolutional filters that process volumetric data across short time windows.
    • cAnalyzing dense, consecutive frames with recurrent neural networks to model temporal dependencies.
    • dRandomly sampling short video snippets from uniformly divided segments across the entire video duration.
    Show the answer

    Answer: d · Randomly sampling short video snippets from uniformly divided segments across the entire video duration.

    The card states that TSN divides a video into segments and samples sparse snippets from each, then aggregates features to understand the entire action. This sparse sampling across the full timeline is how it captures long-range context. Option B describes 3D CNNs, a different architecture that processes local spatio-temporal cubes, not the global sparse sampling of TSN.

    Read the full bite: Temporal Segment Networks: Seeing the Whole Video Story

30. Question 30 of 30

    What is the primary function of the "shifted window" mechanism in Swin Transformers?

    • aTo reduce the computational complexity of self-attention to be linear with image size.
    • bTo enable the application of global self-attention across the entire image more efficiently.
    • cTo ensure that all image patches are processed within a single, fixed window size.
    • dTo allow information exchange between adjacent local windows in successive layers.
    Show the answer

    Answer: d · To allow information exchange between adjacent local windows in successive layers.

    The shifted window mechanism's core purpose is to facilitate information flow between isolated local windows across different layers, building a global understanding. While Swin Transformers achieve linear computational cost, this is primarily due to the initial local window attention, not the shifting mechanism itself.

    Read the full bite: Swin Transformer: Efficient Vision with Shifted Windows