Interview questions in Computer Vision, page 2
SIFT versus SURF versus ORB
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.
Descriptor matching and Lowe's ratio test
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.
Bag of Visual Words model
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.
CNN features for image retrieval
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.
Feature choice for real-time mobile SLAM
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.
Epipolar constraint for correspondence search
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.
Disparity and depth in stereo vision
Disparity is the horizontal shift of a point between left and right images; depth is inversely proportional to disparity given baseline and focal length.
Fundamental matrix versus essential matrix
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.
Incremental Structure from Motion pipeline
Detect and match features, estimate two-view geometry, triangulate, then incrementally add images with PnP and refine via bundle adjustment.
The PnP problem in Structure from Motion
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.
Stereo rectification math and its artifacts
Rectification warps both images by homographies so epipolar lines become horizontal and aligned.
How a convolutional layer works
Learnable kernels slide over the input computing dot products, with stride controlling step size and padding controlling output size.
Max pooling versus strided convolution
Pooling downsamples and adds small translation invariance with no parameters; strided conv learns its downsampling but adds parameters.
Why CNNs need nonlinear activations like ReLU
ReLU introduces nonlinearity letting stacked layers model complex functions; without it any stack collapses to a single linear map.
Regularization techniques for an overfitting CNN
Data augmentation expands the effective dataset, dropout prevents co-adaptation, weight decay penalizes large weights, plus early stopping and batchnorm.
ResNet residual blocks and the degradation problem
A residual block learns F(x) and adds the identity input x, so layers fit a residual; this eases gradient flow and solves the degradation problem.
Receptive fields in convolutional networks
Receptive field is the input region affecting a neuron; it grows with depth, larger kernels, and stride. It matters for capturing context in detection and segmentation.
Uses of the 1x1 convolution
A 1x1 conv is a per-pixel linear combination across channels; it reshapes channel depth cheaply and adds nonlinearity. Uses: dimensionality reduction in bottlenecks and channel mixing.
Adapting a classification CNN for segmentation
Replace the dense head with conv layers, upsample via transposed convolutions, and fuse encoder skip connections to recover spatial detail lost to downsampling.
Depthwise separable convolution cost savings
Separable conv splits standard conv into per-channel spatial filtering plus a 1x1 pointwise mix, cutting cost by roughly 1/N plus 1/k².
We are hiring for this. Every open role lists the topics its interview covers, so you can prepare for the real thing rather than guessing.
See open roles