Datature - a no-code AI platform - simplifies the development and deployment of computer vision solutions for our users. Users can now use our drag-and-drop interfaces to build vision models that typically take hundreds lines of machine learning code, without a single line of code.

For a recap of what computer vision is about, what a typical deep learning pipeline looks like, and what types of problems can be solved using computer vision, click here.

The Pipeline

1. Creating a Project
   The first step of the pipeline is to create a project. You can always make changes to your project after it has been created. Reference Managing your Projects to see how to do that. If you want to work with others on the same project, check out Managing Project Collaborators to add and remove collaborators.

2. Onboarding Your Data and Labels
   The next step of the pipeline is onboarding data and labels to your project. Don't worry if you don't have all your data or annotations collected yet. They can always be uploaded later, through Uploading Data and Uploading Annotations. If you need to retrieve data or annotations uploaded to the platform, you can always download your images again and export annotations.

3. Annotating Your Data
   Often times, data annotation can be a difficult and painstaking process. If you are missing annotations on your images and want to add them, Creating Annotations is easy and adaptable for any of your annotation needs. See the full list of extensive Annotation Tools here. IntelliBrush is the easiest way to get high quality annotations with just a few clicks.

4. Examining Your Annotated Dataset
   To analyze your annotated dataset to see whether it suits your needs and is representative of the tasks that you expect the model to perform, you might want to look at specific subsets of your data. To look at filtered data, whether that's seeing what images have been annotated or images with only certain labels, look at Managing Tags for all those options.

5. Creating Your Training Workflow
   Once you are happy with your dataset, you can begin Creating Training Workflow. A typical workflow requires at least a Dataset module and a Model module. An Augmentations module is optional, but it is recommended. Be sure to check out Module : Dataset , Module : Augmentations , and Module : Model to see what the available options for each module are. However, if you feel that there are too many choices, you can always use the Auto-Suggest Workflow button on the bottom left corner to get started.

6. Previewing Your Augmentations
   If you have added an augmentation module to your workflow, you can check how your augmentations look like by clicking the Preview Augmentations button on the bottom right corner of the workflow page - prior to running the training.

7. Running the Workflow
   Running the Workflow can be initiated by clicking the green Run Training button on the bottom right corner of the workflow page. The Hardware Acceleration, Checkpoint Strategy, and Advanced Evaluation settings are pre-defined for you. However, if in any case you want to specify these settings, check out Hardware Options to find out more about the various options you have for your training.

8. Monitoring Your Training
   Once training has begun, you can start Monitoring Training Process through the trainings page. Our platform provides clear training statuses and error alerts to let you know how the training process is going.

9. Evaluating Model Performance
   When training is over, Evaluating Model Performance is a good way to see if you are happy with the trained model. If you aren't, there are many ways to go about Improving Model Performance.

10. Testing and Using Your Trained Model
    Once you are happy with your trained model, test your model predictions with our no-code application Portal, start using your model through our API, or download your trained models for your own custom deployments. For more information on what to do after a model is trained, head to Managing Trained Models to find out more.