4.2.1 Using Image Files
1. Create a Task:
Create a task for a new training task from a stepper. Select either the existing project or create a new project for creating a task. Refer the image for the flow of creating a training task.
2 Upload Imageset
Upload the images by selecting from the local computer. Either create a new imageset and upload the new images in it or use the existing imageset from the list provided.
Note: For training tasks, minimum 20 Images having 9MP or lower resolution are required
For training task, images can be uploaded in the form of .jpg, .png, .jpeg or the user can upload an orthomosaic map generated by DroneNaksha.
The selection of maximum images for training with precise annotation will improve and assure the accuracy and efficiency of the identification of objects.
The user may set the severity for accuracy from the drop-down menu from "Severity" option. The accuracy levels available are as follows:
Low (Faster)
Medium (Slower)
High (Slowest)
The user can select any one option for severity for accuracy depending on the requirement of the application.
3 Annotate Images:
Annotate images for the objects present in the images. At least the user should annotate 20 images to get better results with better accuracy. This is manual annotation process.
After adding the classes, the user will start the process of annotation. Try to annotate as maximum as possible to get the better results and accuracy.
Select: to select the image
Annotate: This icon allows you to annotate the objects from the image. Select this icon and drag the portion on the image to annotate as shown in the following image.
The user can edit the annotations as shown in the following image. Select an annotation to edit. Select any one option from the displayed options such as:
Copy the annotation block to use it for annotating other objects. The copied annotation can be easily pasted on the image. After pasting the annotation, the user can edit the shape with the edit option.
Delete the unwanted annotations from the image.
Edit the annotations to change the size of an annotation.
Centering: This icon helps to rearrange and display the image in the center of the screen. This type of feature is useful to rearrange the image at the center after zooming in the image.
Complete the annotation of all images and then click on the "Done" button to proceed with the next process.
Note: Annotation will be Input and Detector will be the Output of a training task.
4 Setup
Click on the "Setup" button to setup for training task. Ensure that all the images are fully annotated.
Select Processing Parameters for accuracy level and select the purpose for training.
PicStork provides 3 options for selecting the accuracy level for training Low with faster processing, Medium with slower and High with slowest processing time. Select the accuracy level according to the requirement of the application.
Select the Purpose of Training as "Counting" and "Defect".
Counting: This type of task is used to count something like counting the palm trees from a farm.
Defect: This type of task is used to find out the defects if any.
The report generated from these two tasks will be separate. (refer to section 6.3 of the same user manual for a detailed discussion of reports).
The system will take some time to complete a training task. The time depends on the availability of the server for processing. If the number of such tasks are there then they will be queued. The status of the task will be "Pending" which will be changed to "Completed" after completing the task. Completion of the task will be intimated through the mail to the user.
The summary of a task will be displayed depending on the type of task. The summary of a task is displayed as follows:
Name of Detector
Created Date
Task Name
Created Date
Imageset
No. of Images
Training Type
Accuracy
Status
Completed at
Class Labels
The time required to complete a training task is comparatively more than the other two tasks. For the accurate identification of objects in an image set, precise annotation and rigorous training are required.
Note: Output produced by a training task will be always a detector.
Now, the user can use this detector for performing any other task.
The number of detectors created will be displayed on the dashboard and updated after the successful creation of a detector. The list of detectors can be viewed by clicking on "View" option.
The objects detected through this detector can be made publicly accessible if it has public access given by the user otherwise it will be treated as a private detector and can be used by the owner or user.
Last updated
