Product Demo

Upload

To upload an unstructured dataset, you can utilize our system's support for various file formats. We accept JPGs, PNGs, PDFs, DOCX, PPTX, and many other file types. This allows you to easily upload diverse types of unstructured data for processing.

Customize

Once you have uploaded your dataset, the next step is customizing. At this stage, ontologists and subject matter experts can contribute their intrinsic expertise by:

Choosing their classes

Add, modify, or delete classes to align with your specific use case.

Adding programmatic labeling functions

Incorporate programmatic labeling functions that automate the labeling process based on specific rules or patterns, tailored to your use case.

Annotate

After creating initial programmatic labeling functions, the next step is to begin annotating. We designed our annotation interface to be as seamless, easy to use and enjoyable as possible, almost like a quizlet for data labeling. The GUI is flexible, able to be modified based upon dataset, document type and task. Here, the data labeler sees a row of text data and the corresponding class prediction. The rows shown to the data labeler are sorted by uncertainty (lowest probability) or volatility (biggest impact) that way the data labeler is annotating points that will make the biggest difference to the model. The data labeler can:

• choose the category of the specific label

• mark keywords or bounding boxes that influenced the labeling decision

• skip the row of data if they are unsure about the data label

Annotate for Classification

The Anote platform offers a user-friendly interface for efficient annotation of data for classification tasks. To highlight or mark important features or move bounding boxes, simply interact with the data and utilize the provided tools. By clicking on the corresponding label in the sidebar, users can easily assign the appropriate classification. Once the annotation is complete, a simple click on the "confirm" button allows for seamless transition to the next annotation, optimizing the annotation workflow.

When you are satisfied with your data labels, you can view specific model metrics as well as download as CSV of your high quality, labeled training dataset.

Download

For classification, the download spreadsheet includes the specified columns in the YOLO format. On press of the download button, Anote and converts the table into a CSV file.

On the labeling dataset, this will look like the following table and images, where one box could be the actual (ground truth) labels, and the other box could be the model predictions:

Image Name	Actual Category Label	Actual x center	Actual y center	Actual width	Actual height	Predicted Category Label	Predicted Probability Score	Predicted x center	Predicted y center	Predicted width	Predicted height
img10.jpg	Shark	2100	1450	3500	2500	Shark	0.97	2050	1450	3000	2800
img4.jpg	Jellyfish	575	400	1100	500	Jellyfish	0.96	540	370	850	500

Multiple Labels per Image

When multiple labels exist for a single image, they are stored in multiple rows of a single table. Each row corresponds to one image category identified, allowing comprehensive coverage for images containing several objects.

Image Name	Actual Category Label	Actual x center	Actual y center	Actual width	Actual height	Predicted Category Label	Predicted Probability Score	Predicted x center	Predicted y center	Predicted width	Predicted height
img3.jpg	Shark Jellyfish Fish	100 300 450	120 320 480	80 150 60	60 100 90	Shark Jellyfish Fish	0.93 0.87 0.91	102 295 452	118 325 475	82 148 58	62 98 88

Anote allows for synchronous downloading of the results, providing you with real-time access to the labeled data. As soon as the annotation process is started, you can obtain the results without delay, adjusting in real time to the annotations made.

Integrate via API

You can take the model trained on the data, and export it via API to make improved object detection predictions for undersea imaging. As you curate and label more data, the models performance will be improved. You can use the developer kit to integrate the model into the end-autonomous system, as an more accurate and improved, domain specific real time object detection algorithm.