Semantic Segmentation

Five questions. Choose one answer for each, then submit to see your score and explanations.

The practice target is 80%. You can retry as often as you like, and every lesson stays accessible.

You can practice as a guest. Sign in before starting an attempt to save your scores.

1. What does semantic segmentation assign?A. Only one class for the entire imageB. A class label to each pixelC. Exactly one bounding box to every imageD. A separate identity to every instance by definition

2. What is a key role of U-Net skip connections?A. Bring encoder features into the decoder to help recover spatial detailB. Replace all training labelsC. Force every feature map to one pixelD. Remove the need for a decoder

3. What information does SegNet commonly reuse for decoder upsampling?A. The full set of test labelsB. Only the final image-wide class scoreC. Random box coordinatesD. Pooling indices from the encoder

4. A model gives one label to a whole road scene, but the application needs a road mask. What output is needed?A. The filename of the training imageB. A list of class names without locationsC. A spatial map of road-versus-background predictionsD. A larger single image-level score only

5. You upsample a coarse map to the original image size. What is safe to conclude?A. It no longer needs validationB. Its dimensions increase, but lost fine details are not automatically recoveredC. Every missing boundary is recovered exactlyD. It becomes the mathematical inverse of any encoder

Review this module from its first lesson →

Semantic Segmentation

Your result