Introduction

Detectors Section Overview

The Detectors section provides an organized view of all trained and untrained detectors. Each detector is displayed as a card containing detailed information such as accuracy, classes, creation date, and training status.

Picture1

Key Features:

  1. Detectors Count: Displays the total number of detectors (e.g., 13 detectors).
    • 💡 Tip: Keep an eye on the number of detectors to easily track your progress as you create more detectors over time.

  2. Add Detector: Use the “+” button to create a new detector.
    • 💡 Tip: When starting out, try creating a few detectors with varied use cases to understand how they perform in different situations.

  3. Detector Cards: Each card provides essential information:
    • Name: The name of the detector (e.g., “potholes,” “Pit-New”).
    • Accuracy: Performance accuracy of the detector (e.g., “32.12%”, “88.57%”).
    • Classes: Categories or classes the detector is trained to identify (e.g., “pit,” “Cauliflower”).
    • Date: The date the detector was created (e.g., “27 Dec 2024”).
    • Status: Indicates whether the detector is completed or requires further training.

  4. Status Indicators:
    • Completed: Detectors that are ready for use.
    • Requires Training: Detectors need further training to be functional.
    • 💡 Tip: Keep your detectors updated by regularly retraining them, especially if they are marked as “Requires Training.”

  5. Pagination: Allows navigation between pages of detectors.
    • 💡 Tip: Use pagination to quickly access and manage larger numbers of detectors. This will help in organizing your work, especially when you have multiple detectors to track.

4.1 Managing Detectors

After creating and training detectors, you can manage them with the following options:
1. Delete Detectors:
  • To remove unwanted detectors, click the three dots on the detector card and select Delete.
Picture2
  • A confirmation dialog will appear. Select Yes to delete the detector or No to cancel.

💡 Tip: Only delete detectors you no longer need, as you can always retrain them later if needed.

2. Retrain Models:
  • If a detector is marked as Requires Training, click on the detector card to retrain it.
  • Upload new data and follow the training steps to retrain the model.
💡 Tip: Retraining is crucial to improve the detector’s accuracy, especially when you notice a decrease in performance over time.
3. Monitor Performance:
  • Regularly monitor accuracy and performance metrics to ensure the detectors continue producing optimal results.

💡 Tip: Periodic checks on performance help identify any drift in detector accuracy, allowing for timely adjustments.

4.2 Train a New Detector

The Detectors section enables you to create and train a new detector using uploaded images. To begin, click the “+” button to open the Train a New Detector dialog.

Steps to Train a New Detector

1. Add a New Detector
  • Click the “+” button in the Detectors tab.

This opens the Train a New Detector dialog.

Picture3
  • Enter the following details:
    • Detector Name: Assign a name to the detector.
    • Description (Optional): Provide an optional description.
  • Click Add to proceed.

A Detector Card will be added to the Detectors screen. Click on the Detector Card to view the Train Detector screen, where you can manage and further train the detector.

Picture4
  • 💡 Tip: Naming your detectors based on their specific use cases will help you keep track of them more easily.
2. Add Images for Training
  • On the Train Detector screen, locate the Images Panel on the right-hand side and click the “+” button next to it.

  • This opens the Select Images for Training Detector screen, displaying:
    • A list of available projects containing images.
    • A search bar for easy project location.
Picture5

💡 Tip: Use diverse image sources for training to improve the model’s ability to generalize across different scenarios.

3. Select Images from Existing Projects
  • Search for the desired project or click the project name containing the necessary images.

Select images by ticking the checkboxes on the left side of the image thumbnails.

Picture6
  • Once selected, click Select to confirm.
The Selected images will now appear in the Images Panel of the Train Detector screen.
Picture7

💡 Tip: Ensure the selected images are representative of the variety of objects or patterns you want the detector to recognize.

4. Define Areas

To mark annotations, you need to define the Training, Testing, and Accuracy Areas. You can do this by selecting the respective controls from the toolbar and marking the areas directly on the image.

Picture8

4.1. Define Training Area

  1. On the Train Detector screen, use the left toolbar to select the Training Area

  2. After selecting the Training Area button from the toolbar, mark the points on the image to define the training area.

  3. Once all points are marked, click on the last point to finalize and mark the training area on the image. The area will be highlighted with a yellow dashed line.

4.2. Define Testing Area

  1. On the Train Detector screen, use the left toolbar to select the Testing Area

  2. After selecting the Testing Area button from the toolbar, mark the points on the image to define the testing area.

  3. Once all points are marked, click on the last point to finalize and mark the testing area on the image. The area will be highlighted with a blue dashed line.

4.3. Define Accuracy Area

  1. On the Train Detector screen, use the left toolbar to select the Accuracy Area

  2. After selecting the Accuracy Area button from the toolbar, mark the points on the image to define the accuracy area.

  3. Once all points are marked, click on the last point to finalize and mark the accuracy area on the image. The area will be highlighted with a green dashed line.

💡 Tip: Clearly define all areas (training, testing, and accuracy) to ensure the model learns to differentiate objects correctly.

5. Classes

What is Class?

A Class is a category or label that the detector uses to identify specific objects or patterns within an image. Each class acts as a reference point for the AI to recognize and distinguish various elements during the training process.

  • Example: In an agricultural field image, classes might include “Cauliflower,” “Weed,” or “Soil.”

Purpose of a Class

  • Classes determine what the detector should focus on and categorize in images.
  • They ensure accurate object detection by associating annotations with meaningful labels.

The Classes section allows you to manage the categories used for annotations. A class represents a specific label assigned to the annotated areas.

Picture9

Create a Class

  • Click the Add Class button to create a new class.
  • This will open the Add New Class dialog.
Picture10
  • Enter the desired class name and click the Add button.
  • The newly created class will appear in the Classes section.

Manage Classes

You can manage existing classes by clicking on the three dots (⋮) next to a class name. The following options are available:

Rename

  • After selecting the Rename option, the Update Class dialog will appear.
  • Enter the new class name in the dialog.
Picture11
  • Click the Update button to apply the changes. This will rename the class.

Delete

  • Choose the Delete option from the menu to remove a class.
  • A confirmation dialog will appear with the message:
    “Delete Classes
    Are you sure you want to delete the selected class(es)?”
  • Options available in the dialog:
    • Cancel: To abort the deletion process.
    • Confirm: To permanently delete the selected class(es)

Note: If the detection process is completed, the option to delete the class will be disabled. In such cases, the class cannot be deleted.

6.   Annotations

What is Annotation?

An Annotation is a marked region within an image linked to a specific class. Annotations provide training data that helps the detector learn and recognize patterns.

  • Example: A rectangular annotation around a cauliflower plant labeled as “Cauliflower” helps the AI understand the visual representation of the class.

Purpose of Annotations

  • Annotations define the regions of interest for the detector, enabling it to learn from specific examples.
  • The more precise and varied the annotations, the better the detector’s performance.
Picture12

Annotate Areas

The Annotations section allows you to mark specific areas in an image for training, testing and accuracy. Each annotation is linked to a class and defines the regions of interest for the detector.

Create Annotations

  • After creating a class, the annotation tool in the left toolbar will be activated.
  • Use the annotation shape tool to mark regions on the image.
  • Available shapes include:
  • Rectangles
  • Circles
  • Polygons
  • Carefully annotate areas corresponding to the selected class to ensure accurate training.

Annotation Rules

  • Each annotated area (training, testing, and accuracy) must contain at least 3 annotations.
  • There should be at least 20 annotations across all areas.

💡 Tip: Ensure your annotations are accurate and consistent across images to improve the detector’s learning process.

Manage Annotations

When an annotation is selected, additional management options will appear:

  1. Copy: Duplicate the selected annotation.
  2. Edit: Modify the shape, size, or position of the annotation.
  3. Delete: Remove the annotation if it is no longer required.
  4. Done: Confirm the annotation is complete by clicking the checkmark icon (✔️).
7.   Train the Detector
  • Once all annotations and settings are complete, click the Train Detector button.
Picture13
  • A confirmation dialog will appear:
    • Picture14
    • Click Yes to proceed.
    • Click No to cancel.
    8.   Training Confirmation

    If training begins successfully, a message will appear: “Detector training started successfully.”

    Picture15
    • Click OK to close the dialog and wait for training to complete.

    💡 Tip: If the training takes longer than expected, check the complexity of the dataset and consider simplifying it for faster iterations.

    Pro Tips for Better Performance

    1. Diverse Training Data:
      • Use a variety of images for comprehensive training, which enhances the detector’s ability to generalize across different scenarios.

    2. Regular Retraining:
      • Periodically retrain your detector with new, updated data to improve accuracy and performance over time.

    3. Remove Irrelevant Data:
      • Eliminate low-quality or unnecessary data that could confuse the model, ensuring it focuses on the most relevant features for detection.

    4. Quality Over Quantity:
      • Focus on the quality of your annotations, ensuring they are precise and consistent across images. A smaller set of well-annotated images is often more effective than a large set of poorly annotated ones.
    Table of Contents
    whatsapp
    Scroll to Top