CNN Object Detection Models 4 — Questions and Answers
Question 1: In YOLOv3, how are predictions made at multiple scales?
- By running the network three separate times at different input resolutions
- By detecting at three different feature map scales using a FPN-like structure (Correct answer)
- By applying three different anchor aspect ratios at a single scale
- By ensembling outputs from three separate YOLO models
Correct answer: By detecting at three different feature map scales using a FPN-like structure
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?
- The probability that a detected region belongs to a specific class
- The confidence that a bounding box contains any object (Correct answer)
- The IoU threshold used during NMS
- The anchor box scale multiplier
Correct answer: The confidence that a bounding box contains any object
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?
- Anchor-free bounding box prediction
- A parallel branch for instance segmentation masks (Correct answer)
- Focal loss for class imbalance
- Depthwise separable convolutions in the backbone
Correct answer: A parallel branch for instance segmentation masks
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?
- Using only positive examples (objects) during training
- Selectively sampling background examples that the model misclassifies as objects (Correct answer)
- Augmenting images with artificially generated hard objects
- Using a high learning rate to force the model past local minima
Correct answer: Selectively sampling background examples that the model misclassifies as objects
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?
- They define the output image resolution
- They serve as reference boxes of predefined scales and aspect ratios for regression (Correct answer)
- They replace the need for a backbone network
- They determine the learning rate for each layer
Correct answer: They serve as reference boxes of predefined scales and aspect ratios for regression
Anchor boxes provide predefined reference shapes that the network adjusts via regression to match actual object locations and sizes.
Question 6: Which backbone architecture is commonly used in modern object detectors like Faster R-CNN and RetinaNet?
- LeNet-5
- ResNet with FPN (Correct answer)
- AlexNet
- VGG-7
Correct answer: ResNet with FPN
ResNet combined with a Feature Pyramid Network is a widely used backbone for object detection due to its strong multi-scale feature representation.
Question 7: What does 'stride' control in the context of the backbone's impact on object detection?
- The number of anchor boxes per location
- The spatial resolution of the output feature map relative to the input (Correct answer)
- The number of output classes
- The learning rate decay schedule
Correct answer: The spatial resolution of the output feature map relative to the input
Stride determines how much the feature map is downsampled; a stride of 32 means each feature map cell corresponds to a 32×32 pixel region in the input.
CNN Object Detection Models 4
In YOLOv3, how are predictions made at multiple scales?