Get Started
Quickstart with BinaExperts
Last updated
Was this helpful?
Quickstart with BinaExperts
Last updated
Was this helpful?
BinaExperts provides everything you need to label, train, and deploy computer vision solutions. This guide will help you build a computer vision model for your specific use case.
Quickstart Tutorial (6 Minutes)
A. Adding Data
After reviewing and accepting the terms of service, you will be prompted to choose one of three plans: the Public Plan, the Starter Plan, or the Custom Plan.
B. Invite Collaborators
Next, you'll be asked to invite collaborators to your workspace. Collaborators can help annotate images or manage vision projects. Once collaborators are invited (if desired), you can create a project.
C. Selecting Types of Computer Vision Projects in BinaExperts
For this example, we'll use a dataset suited for quality assurance in a manufacturing facility. You can use any images you prefer to train your model.
Leave the project type as "Object Detection" since our model will identify specific objects and their locations within an image.
Click Create Project to continue.
For this walkthrough, we’ll use the helmet dataset provided by BinaExperts. Download the “helmet” dataset.
Once you've downloaded the dataset, unzip the file. Drag the helmet-dataset folder from your local machine to the highlighted upload area. This dataset is structured in COCO JSON format, one of the 5+ computer vision formats supported by BinaExperts.
To start, we recommend using 5-10 images that balance the classes you want to identify. For example, if you need to identify helmet, ensure there are at least 5-10 images featuring object.
When you upload the helmet-dataset folder, BinaExperts processes the images and annotations for you to view them overlaid.
You can use this guide to build a computer vision model for your unique use case.
BinaExperts will alert you to any annotation errors. For example, if an annotation improperly extends beyond the image frame, BinaExperts will intelligently crop the annotation to the image's edge and remove erroneous annotations outside the frame.
At this stage, your images haven't been uploaded yet. Ensure all images are correct and that annotations are parsed properly. You can delete any image by hovering over it and selecting the trash icon.
If any image is marked "Not Annotated," it can be annotated in the next section. Click Save and Continue to upload your data.
Decide how to split your images among the Train, Test, and Validation sets.
Train Set: Used to train your model.
Validation Set: Used to validate model performance during training.
Test Set: Used to test model performance manually.
Choose the "Use existing values" option to use the pre-split dataset provided by BinaExperts.
You can upload videos in addition to images. Go to the Upload tab, drag a video or paste a YouTube URL. You'll be asked how many frames per second should be captured. For example, choosing " 1 frame/second" will capture an image every second of the video.
Use BinaExperts Annotate to add annotations, such as bounding boxes, around unlabeled objects. These annotations teach your model by serving as an answer key. You can annotate more images to improve model learning. To draw a box, use your cursor to select the area on the image you want to annotate, and then assign a label to it.
Pro Tip: Use BinaExperts AutoLabel to utilize previous model versions for annotating future datasets. AutoLabel leverages an existing model to create annotations automatically, accelerating the process.
BinaExperts also supports polygon annotations, which are necessary for projects requiring precise object localization. To use Polygon, click the sparkles icon in the sidebar.
To add images or data to the dataset, click the Add New Data option.
Once annotations are complete, generate a new dataset version. This snapshot includes images processed in a specific way, similar to version control for data.
You can apply preprocessing and augmentations, such as resizing, grayscaling, random flips, rotations, brightness adjustments, blurring, shearing, cropping, and more. Begin with one or two augmentations and adjust as needed. Generally, avoid augmentations in the first training job to assess the model's performance.
Tip: Some augmentations may not suit your data. Consider the specific use case before applying them. For example, if the object appears only in one orientation (like on an assembly line), flipping may not be helpful.
Click Create to generate your dataset, and it will be ready for use in a few moments.
To train a model on the BinaExperts platform:
I. Click the Train button.
II. Configure your model training job.
III. Choose a model type. For this guide, select 'Fast.'
IV. Select the checkpoint for training.
V. Monitor training progress through real-time graphs.
You'll receive an email with training results once training is complete.
Your trained model is optimized and ready for deployment. BinaExperts supports various deployment options, such as API integration or edge device deployment.
With BinaExperts Hosted API (Remote Server), the model runs in the cloud, eliminating the need to worry about device hardware capabilities.
You can also deploy models using other options suitable for your production applications. For details, refer to our deployment guide.
