More in Computer Vision — page 6
Regularization techniques for an overfitting CNN
WHAT IT TESTS: practical remedies for overfitting and their mechanisms. OUTLINE: data augmentation expands the effective dataset, dropout prevents co-adaptation, weight decay penalizes large weights, plus early stopping and batchnorm.
Why CNNs need nonlinear activations like ReLU
WHAT IT TESTS: why nonlinearity matters in deep networks. OUTLINE: ReLU introduces nonlinearity letting stacked layers model complex functions; without it any stack collapses to a single linear map.
Max pooling versus strided convolution
WHAT IT TESTS: downsampling tradeoffs in CNNs. OUTLINE: pooling downsamples and adds small translation invariance with no parameters; strided conv learns its downsampling but adds parameters.
How a convolutional layer works
WHAT IT TESTS: the mechanics of convolution. OUTLINE: learnable kernels slide over the input computing dot products, with stride controlling step size and padding controlling output size.
Stereo rectification math and its artifacts
WHAT IT TESTS: epipolar geometry and stereo correspondence. OUTLINE: rectification warps both images by homographies so epipolar lines become horizontal and aligned.
The PnP problem in Structure from Motion
WHAT IT TESTS: geometric camera pose estimation. OUTLINE: PnP recovers a camera's pose from known 3D points and their 2D projections; it registers new frames against the existing point cloud in SfM. RED FLAG: confusing it with triangulation.
Incremental Structure from Motion pipeline
WHAT IT TESTS: 3D reconstruction workflow. OUTLINE: detect and match features, estimate two-view geometry, triangulate, then incrementally add images with PnP and refine via bundle adjustment.
Fundamental matrix versus essential matrix
WHAT IT TESTS: epipolar algebra. OUTLINE: both relate corresponding points across two views via the epipolar constraint; the fundamental matrix works in pixel coordinates, the essential matrix in calibrated coordinates and encodes relative pose.
Disparity and depth in stereo vision
WHAT IT TESTS: stereo depth fundamentals. OUTLINE: disparity is the horizontal shift of a point between left and right images; depth is inversely proportional to disparity given baseline and focal length.
Epipolar constraint for correspondence search
WHAT IT TESTS: two-view geometry basics. OUTLINE: the match for p1 must lie on its epipolar line in the second image, reducing a 2D search to 1D; the relation is encoded by the fundamental matrix.
Feature choice for real-time mobile SLAM
WHAT IT TESTS: system-level feature trade-offs. OUTLINE: pick ORB for fast FAST keypoints and cheap binary descriptors matched by Hamming distance; accept reduced robustness versus SIFT for real-time, low-power operation.
CNN features for image retrieval
WHAT IT TESTS: transfer learning for features. OUTLINE: pass the image through a pretrained CNN and read activations from a late layer as a descriptor; deeper layers encode semantics, earlier layers encode texture.
Bag of Visual Words model
WHAT IT TESTS: classic image representation. OUTLINE: cluster many local descriptors (e.g. k-means) into visual words; assign each image's features to words; represent the image as a histogram of word counts for a classifier.
Descriptor matching and Lowe's ratio test
WHAT IT TESTS: robust feature matching. OUTLINE: match by nearest-neighbor descriptor distance; Lowe's ratio test keeps a match only if the best is clearly better than the second-best, rejecting ambiguous ones.
SIFT versus SURF versus ORB
WHAT IT TESTS: feature-method trade-offs. OUTLINE: SIFT is most accurate but slow with float descriptors; SURF approximates SIFT for speed; ORB is fast, binary, and free, ideal for real-time and embedded.
SIFT scale and rotation invariance
WHAT IT TESTS: how SIFT achieves invariance. OUTLINE: scale-space extrema via difference-of-Gaussians give scale invariance; a dominant gradient orientation gives rotation invariance; the descriptor is a normalized gradient histogram.
Image gradients, Sobel, and Canny
WHAT IT TESTS: edge-detection foundations. OUTLINE: the gradient measures local intensity change in x and y; Sobel approximates it via convolution kernels; Canny uses gradient magnitude and direction plus non-max suppression and hysteresis.
Harris corner detector and corner stability
WHAT IT TESTS: feature detection intuition. OUTLINE: Harris finds points where intensity changes strongly in all directions using the structure tensor of gradients; corners are well localized in two directions, unlike edges.
Image rotation: forward versus inverse mapping
WHAT IT TESTS: geometric warping mechanics. OUTLINE: forward mapping sends source pixels to non-integer destinations, leaving holes and overlaps; inverse mapping iterates over output pixels, finds the source location, and interpolates.
Removing salt-and-pepper noise
WHAT IT TESTS: filter choice for impulse noise. OUTLINE: use a median filter; it replaces a pixel with the neighborhood median so extreme outliers are discarded. RED FLAG: choosing Gaussian blur, which averages outliers in and smears edges.