Glossaire

Détecteurs d'objets à deux niveaux

Discover the precision of two-stage object detectors in computer vision, ideal for tasks demanding high accuracy in autonomous vehicles and healthcare imaging.

Train YOLO models simply
with Ultralytics HUB

En savoir plus

Two-stage object detectors are a prominent approach in the field of computer vision, particularly known for their accuracy in tasks like detecting and identifying various objects within an image. These systems operate by breaking down the detection process into two sequential stages, offering a detailed and robust method for identifying objects with higher precision compared to one-stage object detectors.

How Two-Stage Object Detectors Work

The process begins with the first stage, which generates potential regions of interest (ROIs) in the image. This stage uses techniques to identify areas that likely contain objects without pinpointing the objects themselves. Common methods include Region Proposal Networks (RPNs) which efficiently provide candidate object locations.

In the second stage, the detector refines these proposals by classifying the identified regions and adjusting their boundaries to better fit the objects. The refinement includes more detailed analysis using a Convolutional Neural Network (CNN) to classify the object and further define its boundaries.

Comparaison avec les détecteurs à un étage

While two-stage detectors are valued for their accuracy, they tend to be slower than one-stage object detectors such as the Ultralytics YOLO family. One-stage detectors skip the ROI proposal phase and make predictions directly over dense sampling of possible object locations. This direct method can be faster but may sacrifice some accuracy, making two-stage detectors preferable for applications where precision is crucial.

Examples of Two-Stage Object Detectors

  • R-CNN and Variants: The original R-CNN (Region-based Convolutional Neural Network) paved the way for faster models like Fast R-CNN and Faster R-CNN, each optimizing speed and accuracy. Faster R-CNN is commonly used in scenarios where accuracy is prioritized, such as medical imaging or autonomous vehicle technology.

  • Mask R-CNN: An extension of Faster R-CNN, Mask R-CNN not only detects objects but also provides a pixel-level mask of each object. It's widely used in cases requiring instance segmentation beyond mere object detection, such as in the fashion industry for automated apparel tagging (Explore Mask R-CNN).

Applications in Real World

Véhicules autonomes

In self-driving cars, two-stage detectors are utilized to identify pedestrians, cyclists, and vehicles with high accuracy, ensuring safety and adherence to road regulations. AI in self-driving vehicles relies heavily on these detectors for their decision-making systems.

Imagerie médicale

Two-stage object detectors are instrumental in analyzing medical images, helping to identify tumors, fractures, or other critical conditions accurately. In healthcare, where precision is vital, these models facilitate better diagnostic processes and outcomes. Vision AI in healthcare showcases various applications transforming the medical field.

Integration and Future Prospects

With advancements in AI and machine learning, two-stage object detectors are becoming increasingly integrated with other technologies like Transfer Learning and AI Ethics. The integration with platforms like Ultralytics HUB allows for seamless training and deployment, making state-of-the-art object detection accessible to a wider audience.

The future of two-stage object detection looks promising with continuous improvements in algorithm efficiency and hardware capabilities. This progress ensures that they remain a fundamental part of AI-driven solutions in various complex domains. For those interested in leveraging these technologies, exploring Ultralytics' resources and solutions can provide substantial support and guidance.

Read all