Single-stage detector

As explained in Deep Learning with R, Third Edition

A single-stage object detector predicts object locations and categories directly in one integrated model pass rather than first generating region proposals.

A single-stage detector predicts object locations and categories together in one model pass. Unlike an R-CNN detector, it does not first generate region proposals and then classify thousands of patches. This integrated design is generally faster and more computationally efficient, making it useful for real-time and resource-constrained applications. section 12.1.1 section 12.1.2

A typical single-stage detector works by:

• Extracting high-level image features with a convolutional backbone. section 12.2.2
• Predicting bounding boxes, confidence scores, and class labels directly from those features. section 12.2.2 section 12.4
• Using a grid or multiple feature scales to associate predictions with different image locations and object sizes. section 12.2.2 section 12.3

A simplified YOLO-style detector

Resize images to 448 × 448, pass them through a pretrained ResNet backbone, and reshape the prediction head into a 6 × 6 grid. Each grid cell outputs one box plus 91 class values; the box output contains coordinates and confidence, while the class output uses softmax probabilities. The simplified model predicts one box per cell, whereas the original YOLO design predicted multiple boxes. section 12.2.2

RetinaNet follows the same single-stage principle but improves detection across object sizes with a Feature Pyramid Network. The FPN combines semantically rich deep features with higher-resolution features from earlier layers, helping detect both small and large objects. section 12.3

Figure 12.8 An FPN creates semantically interesting feature maps at different scales.

Shows how an FPN creates semantically meaningful feature maps at multiple spatial scales. section 12.3

How Deep Learning with R, Third Edition covers it

How deep it goes

Explains proposal avoidance, direct outputs, grid prediction, and multi-scale features, while simplifying YOLO to one box per grid cell.

Examples

Uses a simplified YOLO grid, pretrained ResNet, 448 × 448 inputs, and RetinaNet inference on a COCO-trained painting example.

What you can do after

Build it from scratch with a convolutional backbone and prediction head, or apply it through a pretrained RetinaNet detector.

How it connects

Connects object detection with bounding boxes, confidence scores, class probabilities, convolutional backbones, YOLO, RetinaNet, R-CNN, and Feature Pyramid Networks.

liveBookmanning.com

Read Deep Learning with R, Third Edition on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Deep Learning with Python, Third EditionAn object-detection architecture that predicts bounding boxes and class labels in one pass with a single model, rather than using separate proposal and classification stages.

Compare how 2 books explain single-stage detector →

Related concepts

• RetinaNetincludes · RetinaNet is a one-stage object-detection architecture that uses convolutional feature pyramids with separate classification and bounding-box regression heads.
• YOLOincludes · YOLO is a family of single-stage object-detection models that predict object classes and locations directly from an image in one inference pass.
• object detection modelkind of · An object detection model is a computer-vision model that identifies object instances and predicts their classes and spatial locations in visual data.
• object detectionaddresses · Object detection is a computer-vision task that identifies object instances in visual data, assigns them categories, and estimates their locations.

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with R, Third Edition. The section links point to the parts of the book it draws on.