The new visage|SDK 8.7 release combines all the improvements and novelties that have been published within the previous two beta releases. It is primarily focused on improvements in face detection, face alignment, and face recognition.
- Smaller, faster, and more accurate face recognition model
The new face recognition model files are significantly smaller, requiring less space for their integration. This is especially important for devices with limited memory, such as smartphones. The smaller model also loads and performs faster, making the face recognition algorithm up to 70% faster. Finally, the accuracy of the model has been improved, reducing the likelihood of false positives and false negatives.
- The improved face detection model
The new face detection model is more robust to various challenging conditions such as faces with high variability in scale (different distances from the camera), illumination, pose and occlusion.
- VNN face tracking algorithm improvements
The new release introduces a fast mode of the VNN algorithm which improves performance at the cost of feature points precision (while keeping the same precision of head pose). The algorithm now also works with up to 40% higher FPS, with significant improvements on devices such as high-end mobile and desktop devices. Finally, the tracking noise on mobile platforms – Android and iOS – is now reduced, resulting in less tracking jitter.
Browse the documentation online
The new release is accompanied by another exciting announcement – visage|SDK documentation is now available online as well.
This makes exploring and getting started with our technology easier and more convenient. Using a simple search option, you can quickly find exactly what you need. Read about interesting use cases, explore samples, find answers to the most common questions, and more in just a few clicks.
We are currently working on many new features and improvements that should be available soon. Some of the novelties you can look forward to include:
- VNN algorithm coming to HTML5 – more precise and more robust to occlusions and lighting conditions, and fast to boot
- Head pose improvements – face contour detection as a way of providing a robust and stable head pose estimation
- Face masks – working hard to bring robust and precise tracking even with face masks
- Virtual makeup SDK – simple addition of various makeup products into virtual makeup try-on solutions