Quiz dotyczący biblioteki widzenia komputerowego OpenCV

Odpowiedzi i wyjaśnienia pojawiają się w miarę rozwiązywania quizu.

1. Which option names operations that directly alter image values or geometry?
   • Estimate local feature descriptors for later matching
   • Decode video frames into image arrays
   • Filtering and geometric image transformations
   • Detect object locations without changing the input pixels
2. In the standard OpenCV image-reading path, which channel order is commonly used for color images?
   • RGB in every API
   • BGR
   • CMYK
   • Grayscale only
3. How does OpenCV commonly relate to PyTorch or TensorFlow in an ML pipeline?
   • It automatically replaces every dedicated framework's training loop
   • Its image I/O removes the need to define preprocessing
   • It requires every model to be trained inside its DNN module
   • It can preprocess data while another framework trains the model
4. What must be checked when handing an OpenCV array to another model runtime?
   • Shape, datatype, range, layout and channel order
   • Checking only array shape is sufficient
   • Checking datatype and range but ignoring channel order is sufficient
   • Assuming every binding returns a contiguous RGB array
5. What determines whether a particular codec or GPU path is available?
   • The package name alone means GPU support is enabled
   • A GPU device guarantees every module was compiled with acceleration
   • The package build and runtime environment
   • File format support is identical in every build
6. What does camera calibration estimate or correct for?
   • The semantic classes present in the scene
   • Camera geometry such as intrinsics and lens distortion
   • The decision threshold for a recognition model
   • Which video codecs were compiled into the package
7. Why should preprocessing match the model's training recipe?
   • Matching resolution makes every other preprocessing difference harmless
   • Normalization can change freely if array datatype stays the same
   • A saved model automatically repairs every input-representation mismatch
   • Different resize or normalization steps can shift predictions
8. Which statement best limits what OpenCV alone provides?
   • A computer vision library, not a complete validated product
   • A complete training and deployment solution for every vision task
   • A guarantee that all preprocessing matches a downloaded model
   • An identical set of hardware backends in every package build