Nilvana Vision Studio - Machine Annotation
Please follow the instructions to get started machine annotation with nilvana vision studio.
Machine Annotation Types Provided by Vision Studio
Data annotation is a time-consuming task in the model-making process. Vision Studio provides a variety of annotation methods, among which machine annotation can automatically help you complete annotation work through the existing models, which not only shortens the time of annotation work, but also reduces the manpower, so that you can focus on more critical areas.
Vision Studio provides the following annotation types:
- Manual annotation, complete annotation work through our real-time manual annotation tool, which is constantly tested and optimized with the best user experience design.
- Machine annotation
- Semi-automatic annotation(Recommended): Loading the built-in or trained models for machine annotation, the system automatically detects the objects and adds the dotted line annotation suggestion box. You only need to check and click for fine-tuning to complete the annotation work.
- Automatic annotation: Loading the built-in or trained models for machine annotation, the system will automatically detect the objects and directly mark the annotation box, just as if you annotated it manually.
Provide Built-In Models for Machine Annotation
When you first use machine annotation, you can start enjoying the convenience of machine annotation directly through the built-in model that we provide. Most of the built-in models are based on open datasets and have been evaluated and trained by us. They include:
- Mask detection model: providing the wearing verification of masks, including wearing masks, not wearing masks, and wearing masks incorrectly.
- COCO 80 model: trained from the COCO (Common Object in Context) dataset, covering over 80 classes of commonly used objects, such as cars.
If these models do not contain the types you need, don't worry! You can follow us further and generate your own machine annotation model through Vision Studio.
Generate Your Own Machine Annotation Model
If your annotation object is not included in the system, you can follow the process below to generate your own simple model, and then use this simple model to complete the remaining annotation work, saving you the time and cost of annotation.
- Annotate at least 500 images for each class using the manual annotation tool.
- Generate your first model with the automatic model training tool that we have carefully prepared for you. If you are not sure how to use this tool, please refer to "Nilvana Vision Studio - Model Training".
- Load this model for machine annotation in order to automatically annotate the remaining large number of images and improve annotation efficiency.
- In the end, you will just need to quickly check the results of machine annotation, without having to annotate them individually.
Adjust Minimum Score Threshold for Machine Annotation
When using machine annotation, you can filter its content by a minimum score threshold. The threshold level will affect the results of machine annotation. The higher the value is, the higher the chances of reducing error, but also the greater the chance of missing some objects. Although the system's default value is 50%, the optimal threshold depends on the data you have. You can also select the classes you want at this point and do not need to apply to all the classes.
Finally, press "Save" to complete the machine annotation. Machine annotation is an auxiliary tool that can do most of the annotation work, but the accuracy of annotation ultimately depends on the model used. Therefore, during the manual inspection process, we can also improve the annotation through the manual annotation function provided when there are unannotated images.