Skip to content
My Docs

42_Object_Detection__Yolo__R Cnn_

Object Detection (YOLO, R-CNN)

Section titled “Object Detection (YOLO, R-CNN)”

Category: Computer Vision
Type: AI/ML Concept
Generated on: 2025-08-26 11:04:24
For: Data Science, Machine Learning & Technical Interviews

Object Detection Cheatsheet (YOLO, R-CNN)

Section titled “Object Detection Cheatsheet (YOLO, R-CNN)”

1. Quick Overview

Section titled “1. Quick Overview”

What is Object Detection?

Object detection is a computer vision task that involves identifying and locating objects within an image or video. It goes beyond simple image classification by not only recognizing what objects are present but also where they are located (bounding boxes).

Why is it important?

  • Automation: Automates tasks that previously required human intervention (e.g., security surveillance, quality control).
  • Contextual Understanding: Provides richer information about scenes, enabling more intelligent systems.
  • Foundation for other tasks: A critical building block for more complex AI systems like autonomous driving, robotics, and video analytics.

2. Key Concepts

Section titled “2. Key Concepts”

  • Bounding Box: A rectangular box that encloses an object of interest. Defined by:

    • (x_min, y_min): Top-left corner coordinates
    • (x_max, y_max): Bottom-right corner coordinates
    • Alternatively: (x_center, y_center, width, height)

  • Intersection over Union (IoU): A metric to evaluate the overlap between two bounding boxes (predicted and ground truth).

    • IoU = Area of Intersection / Area of Union
    • A higher IoU indicates a better prediction. Typically, a threshold of 0.5 or higher is used to consider a prediction as a true positive.

  • Ground Truth: The actual bounding box and class label of an object in the image, manually annotated.

  • Confidence Score: A measure of how confident the model is that an object is present within a predicted bounding box.

  • Non-Maximum Suppression (NMS): A technique to eliminate redundant bounding boxes that are highly overlapping and predicting the same object. It keeps the bounding box with the highest confidence score.

  • Feature Extraction: The process of extracting relevant features from an image using techniques like convolutional neural networks (CNNs).

  • Region Proposal: Algorithms (e.g., Selective Search) that identify potential regions in the image that might contain objects.

  • Anchor Boxes: Predefined bounding boxes of different shapes and sizes used in YOLO to predict object locations and dimensions relative to these anchors.

3. How It Works

Section titled “3. How It Works”

3.1. R-CNN (Regions with CNN features)

Section titled “3.1. R-CNN (Regions with CNN features)”

  1. Region Proposal: Selective Search algorithm generates a set of region proposals (around 2000) that are likely to contain objects.

    +-----------------+
    | Image           |
    +-----------------+
         |
         V
    +-----------------+
    | Selective Search|
    +-----------------+
         |
         V
    +-----------------+
    | Region Proposals| ( ~2000 )
    +-----------------+

  2. Feature Extraction: Each region proposal is warped to a fixed size and passed through a CNN (e.g., AlexNet) to extract a feature vector.

    +-----------------+
    | Region Proposal |
    +-----------------+
         |
         V
    +-----------------+
    | CNN (AlexNet)   |
    +-----------------+
         |
         V
    +-----------------+
    | Feature Vector  | (4096 dimensions)
    +-----------------+

  3. Classification: Each feature vector is fed into a set of Support Vector Machines (SVMs) to classify the object within the region proposal. One SVM is trained for each object class.

    +-----------------+
    | Feature Vector  |
    +-----------------+
         |
         V
    +-----------------+
    | SVM Classifiers | (one per class)
    +-----------------+
         |
         V
    +-----------------+
    | Class Probabilities|
    +-----------------+

  4. Bounding Box Regression: A linear regression model is used to refine the bounding box coordinates to better fit the object.

    Improvements over R-CNN: Fast R-CNN, Faster R-CNN. Faster R-CNN replaces Selective Search with a Region Proposal Network (RPN) within the CNN, making it significantly faster.

3.2. YOLO (You Only Look Once)

Section titled “3.2. YOLO (You Only Look Once)”

YOLO is a one-stage object detector that directly predicts bounding boxes and class probabilities from the entire image in a single pass.

  1. Image Division: The input image is divided into an S x S grid.

    +-----------------+
    | Input Image     |
    +-----------------+
         |
         V
    +-----------------+
    | S x S Grid      | (e.g., 7x7)
    +-----------------+

  2. Bounding Box Prediction: Each grid cell is responsible for predicting B bounding boxes. For each bounding box, it predicts:

    • (x, y): Center coordinates of the bounding box relative to the grid cell.
    • (w, h): Width and height of the bounding box relative to the entire image.
    • Confidence Score: The probability that an object exists in the bounding box and how accurate the bounding box is.
    • C Class probabilities (one probability per class).
      • Each bounding box predicts confidence score and coordinates. Each grid cell predicts class probabilities.

  3. Non-Maximum Suppression (NMS): NMS is applied to filter out redundant bounding boxes with lower confidence scores.

YOLO Architecture (Simplified):

+-----------------+
| Input Image     |
+-----------------+
     |
     V
+-----------------+
| Convolutional   |
| Neural Network  | (Feature Extraction)
+-----------------+
     |
     V
+-----------------+
| Fully Connected |
| Layers          | (Prediction)
+-----------------+
     |
     V
+-----------------+
| Output: Bounding|
| Boxes & Classes |
+-----------------+

YOLO Output Tensor:

The output of YOLO is a tensor of shape (S, S, B * (5 + C)), where:

  • S: Number of grid cells along each dimension.
  • B: Number of bounding boxes predicted by each grid cell.
  • 5: Represents the 4 bounding box coordinates (x, y, w, h) and the confidence score.
  • C: Number of classes.

Example (YOLOv3):

If S=13, B=3, and C=80 (for COCO dataset), the output tensor shape would be (13, 13, 3 * (5 + 80)) = (13, 13, 255).

4. Real-World Applications

Section titled “4. Real-World Applications”
  • Autonomous Driving: Detecting vehicles, pedestrians, traffic signs, and lane markings.
  • Security Surveillance: Identifying suspicious activities, detecting intruders, and tracking objects.
  • Retail: Counting customers, analyzing shopping behavior, and detecting shoplifting.
  • Manufacturing: Quality control, defect detection, and robotic assembly.
  • Medical Imaging: Detecting tumors, identifying anomalies, and assisting in diagnosis.
  • Agriculture: Detecting diseases in crops, monitoring livestock, and optimizing irrigation.

5. Strengths and Weaknesses

Section titled “5. Strengths and Weaknesses”

R-CNN Family (e.g., Faster R-CNN):

  • Strengths:
    • High accuracy, especially with Faster R-CNN.
    • Well-established and widely used.
  • Weaknesses:
    • Slower than YOLO, especially the original R-CNN.
    • More complex to implement and train.

YOLO:

  • Strengths:
    • Very fast and suitable for real-time applications.
    • Simpler architecture compared to R-CNN family.
    • Learns generalizable representations of objects.
  • Weaknesses:
    • Lower accuracy than Faster R-CNN, especially for small objects or objects in dense scenes.
    • Can struggle with objects that are close together.

6. Interview Questions

Section titled “6. Interview Questions”

Q: What is object detection?

A: Object detection is a computer vision task that involves identifying and locating objects within an image or video. It aims to determine what objects are present and where they are located, typically by drawing bounding boxes around them.

Q: Explain the difference between image classification and object detection.

A: Image classification assigns a single label to an entire image, while object detection identifies and localizes multiple objects within an image, providing bounding boxes and class labels for each object.

Q: What is Intersection over Union (IoU)? How is it used in object detection?

A: IoU is a metric that measures the overlap between two bounding boxes. It’s calculated as the area of intersection divided by the area of union. In object detection, IoU is used to evaluate the accuracy of predicted bounding boxes compared to ground truth bounding boxes. A high IoU indicates a better prediction.

Q: Explain Non-Maximum Suppression (NMS).

A: NMS is a post-processing technique used to eliminate redundant bounding boxes that are predicting the same object. It works by sorting bounding boxes by their confidence scores and iteratively suppressing boxes that have a high IoU with a higher-scoring box.

Q: Compare and contrast R-CNN and YOLO.

A: R-CNN is a two-stage detector that first proposes regions of interest and then classifies them. It is generally more accurate but slower than YOLO. YOLO is a one-stage detector that directly predicts bounding boxes and class probabilities from the entire image in a single pass. It is much faster but can be less accurate, especially for small objects.

Q: What are anchor boxes and why are they used in YOLO?

A: Anchor boxes are predefined bounding boxes of different shapes and sizes. They are used in YOLO to predict object locations and dimensions relative to these anchors, allowing the model to handle objects with varying aspect ratios and scales more effectively.

Q: What are some real-world applications of object detection?

A: Autonomous driving, security surveillance, retail analytics, manufacturing quality control, medical imaging, and agricultural monitoring.

Q: How does Faster R-CNN improve upon R-CNN?

A: Faster R-CNN replaces the Selective Search algorithm used in R-CNN with a Region Proposal Network (RPN) within the CNN. This RPN learns to generate region proposals, making the process much faster and more efficient.

7. Further Reading

Section titled “7. Further Reading”