CNN Object Detection Models 4

Question 1

In YOLOv3, how are predictions made at multiple scales?

Accepted Answer

By detecting at three different feature map scales using a FPN-like structure

Answer

YOLOv3 predicts bounding boxes at three scales by extracting feature maps at different depths, similar to a feature pyramid network.

Question 2

What is the 'objectness score' in YOLO detections?

Accepted Answer

The confidence that a bounding box contains any object

Answer

The objectness score estimates the probability that the bounding box actually contains an object (regardless of class).

Question 3

Which innovation did Mask R-CNN add to Faster R-CNN?

Accepted Answer

A parallel branch for instance segmentation masks

Answer

Mask R-CNN adds a small FCN branch that predicts a binary segmentation mask for each detected object instance in parallel with the class and box heads.

Question 4

What does the term 'hard negative mining' mean in object detection training?

Accepted Answer

Selectively sampling background examples that the model misclassifies as objects

Answer

Hard negative mining selects the most confusing background regions for training to address class imbalance and improve discrimination.

Question 5

What is the role of anchor boxes in SSD and Faster R-CNN?

Accepted Answer

They serve as reference boxes of predefined scales and aspect ratios for regression

Answer

Anchor boxes provide predefined reference shapes that the network adjusts via regression to match actual object locations and sizes.

CNN - Convolutional Neural Networks Practice Test

CNN - Convolutional Neural Networks Practice Test

CNN Object Detection Models 4