Could you elaborate on the process of pre-training models within PicStork and the procedure for adding objects for identification?
1 Like
Hi @Anand_Nayak,
In the Picstork, there are pre-trained models available for detecting objects in images and orthomosaic maps. You simply need to select them, and the system will identify your chosen objects if they are present in the machine learning model.
If you want to add a new object for identification, you’ll first need to create a training task and develop a machine-learning model according to your specific requirements. Once created, you can then use this model for object identification in your task.
Cheers!
2 Likes
@Anand_Nayak Here are the steps for training the model:
1. Create a project and task for model training.
2. Establish an image set and upload images. If you have an orthomosaic map created on the DroneNaksha, you can also import it into the image set.
3. Initiate the task by annotating the required objects. For image-based training tasks, annotate a minimum of 20 images. For orthomosaic map tasks, draw the Training area, Testing area, and Accuracy area.
4. Proceed to the setup option, where you will select task accuracy and purpose, and start the training task.
5. Once the model is ready, you will receive an email notification.
6. Verify your created model in the detector section of the Picstrok dashboard.
I hope this helps! Let me know if you need further clarification on any of the steps.
2 Likes
Hey, @Anand_Nayak go through this end-to-end video for the training procedure and using the detector for object identification.
