Object detection

As explained in Inside Deep Learning: Math, Algorithms, Models

A vision task in which a system identifies and localizes multiple objects in a single image, often producing several predictions per image.

Object detection identifies what objects appear in one image and where they are. Unlike image classification, it can produce zero, one, or many predictions per image. section 8.0

A practical detector usually predicts a class and a bounding box for each object. A box is represented by four absolute coordinates, such as (x1, y1, x2, y2), and should tightly contain the entire object without duplicating it. section 8.4 section 8.4.2

Two common ways to obtain object locations are:

• Image segmentation classifies every pixel as an object category or background, then connected regions can reveal each object. It is precise but requires expensive pixel-level labels. section 8.1
• Bounding-box detection finds candidate locations in one pass and classifies the objects found there. Faster R-CNN combines a convolutional backbone with networks for proposing regions and deciding their classes. section 8.4 section 8.7
• Non-maximum suppression reduces false positives by filtering overlapping predictions. section 8.7

A minimal PyTorch detection dataset

A toy MNIST detector places a random number of digits on a 100 × 100 canvas. Each sample returns an image and a dictionary: boxes has shape (k, 4), while labels has shape (k), allowing k to vary between images. Faster R-CNN receives lists of images and dictionaries through a custom collate function rather than a padded tensor batch. section 8.4.2

Figure 8.8 A stop sign as the target for bounding-box-based object detection. The green box shows the ground truth, a box just large enough to contain the entire object. The red box shows a potential prediction, which is close but not quite correct. Water doesn’t obey the whimsical desires of the lowly stop sign.

Use the stop-sign diagram to compare a ground-truth box with a nearby but imperfect prediction. section 8.4

How Inside Deep Learning: Math, Algorithms, Models covers it

How deep it goes

Descends from task definition to segmentation, bounding boxes, Faster R-CNN components, target formats, batching, training, and inference.

Examples

Uses birds, horses, people, cars, stop signs, and synthetic MNIST digits to illustrate detection targets and predictions.

What it assumes

Assumes familiarity with image classification and CNNs while teaching variable object counts, box coordinates, and detector-specific batches.

What you can do after

Build it from scratch with PyTorch data targets, custom collation, Faster R-CNN training, inference, and plotted predictions.

How it connects

Connects image classification, image segmentation, convolutional neural networks, bounding boxes, Faster R-CNN, and non-maximum suppression.

liveBookmanning.com

Read Inside Deep Learning: Math, Algorithms, Models on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Machine Learning Platform EngineeringObject detection is a computer-vision task that identifies objects in an image or other visual input and locates them with bounding regions.
• 

  liveBookmanning.com
  Hugging Face in ActionObject detection is the machine learning task of identifying and locating objects within images or video.
• 

  liveBookmanning.com
  Designing Deep Learning Systems: A guide for software engineersObject detection is the computer-vision task of locating instances of target object classes in images and assigning labels to them.
• 

  liveBookmanning.com
  Math and Architectures of Deep LearningA computer-vision task of identifying object categories in an image and estimating their spatial locations, often with bounding boxes.

Compare how 5 books explain object detection →

Related concepts

• image segmentationcontrasts with · Image segmentation is a computer-vision task that partitions an image and assigns labels to its pixels or regions to delineate objects or meaningful areas.
• computer visionpart of · Computer vision is the field of computing concerned with using computational methods to extract and interpret information from images, video, and other visual data.
• bounding boxuses · A bounding box is an axis-aligned rectangular or hyperrectangular region defined by coordinate limits that encloses or localizes an object, area, or set of points.
• confidencemeasured by · Confidence is a degree of certainty or assurance associated with a judgment, estimate, prediction, capability, or statistical procedure, often expressed as a score, probability, or coverage level.

This explainer was generated by Manning's knowledge graph from the text of Inside Deep Learning: Math, Algorithms, Models. The section links point to the parts of the book it draws on.