BasicAI Cloud's latest update brings a suite of new features to your data annotation toolbox. With the new version v0.10.1, we've rolled out a brand-new pixel-level image segmentation mode, automated segmentation incorporating models, and LaTeX expression tools. We've also fine-tuned other image annotation tools, such as shared edges, circle and ellipse drawing for even greater efficiency gains. Read on for details about how these upgrades will enable faster, higher quality data annotation work for your team.

Introducing Pixel-Level Image Segmentation

Electron microscope imaging: fibers and glass bubbles

A major challenge in AI model development today is acquiring real-world data and matching ground truth labels—a time-intensive and costly process. Data quality underpins many machine learning and deep learning tasks, with incorrect annotations potentially causing unstable model training, performance decline, or even non-convergence. Without a human-in-the-loop (HITL) approach, balancing efficiency and precision proves challenging.

While Zero-Shot image segmentation models can recognize distinct objects, manual handling is required when dealing with object boundaries and occluded areas to ensure accuracy.

Image Segmentation Demo

In our 0.10.1 update, we've split image segmentation annotation into two distinct modes: polygon mode and brush mode. The latter integrates seamlessly with AI preprocessing models as it allows pixel-level adjustment of subtle positions, a valuable asset when model outputs are inaccurate.

Brush Tool

Shape: Free-draw lines, outlines, or masks

Large Area: Offers high precision, though at a slower speed

Small Area: Maintains high precision with quicker adjustment speed

Use Cases: Suitable for irregular shapes, doodles, and free shape annotations

Polygon Tool

Shape: Facilitates regular polygon shape creation

Large Area: Fast but possibly less precise for highly irregular targets

Small Area: More error-prone, with challenging adjustments

Use Cases: Ideal for regular boundaries, object contour annotations

LaTeX Support for Formulas

LaTeX is a typesetting system that allows complex tables and mathematical formulas to be generated by keyboard input. This makes it perfectly suited for producing high-quality technical, mathematical, and scientific documents.

After annotating a target with a bounding box, LaTeX code can be entered to render the formula in real-time. Invalid code is called out to avoid errors.

The LaTeX input proves highly practical when transcribing books or handwritten documents.

As Gartner states in their 2023 AI Hype Cycle, data and annotation will have profound impacts on the AI industry:

• Enabling previously impossible AI solutions by addressing scarcity of training data

• Increasing accuracy of foundational models through reinforcement learning with human feedback (RLHF)

• Tailoring outputs of generative AI to meet specific organizational needs

• Boosting performance of AI solutions thanks to larger, annotated datasets

• Accelerating model development and ability to adapt to diverse workloads

For tasks where precision and safety are critical — such as medical image segmentation or road segmentation in autonomous driving — incorrect decisions can lead to severe consequences. Hence, manual annotation is crucial in ensuring data accuracy and credibility, and model-assisted tools play a significant role in expediting this process.