Library models
Scailable maintains a large library of AI models useful for various applications. Currently we support, amongst others:
- Object classification, detection, and localization of common objects such as people, faces, cars, trucks, bikes, and animals.
- Object counting and line-crossing for the objects listed above (e.g., allowing for entrance control and parking lot control).
- Highly accurate Automatic License Plate Recognition (ANPR).
- Motion detection.
- Emotion detection and face localization.
- Highly accurate product quality inspection (given sufficient sample products).
Currently, a small subset of our library models is available directly from the AI manager. Additional library models are available for licensed users upon request.
Many library models contain named input and output options that provide contextual pre- and post-processing.
The table below contains models available to be used off-the-shelf.
Most of these models have two different versions based on the input resolution. This allows the user to trade-off between a model's speed and its performance.
For example, the people locator (first and second row in the table below) have the 128 x128 and 256 x256 input resolutions. The former is relatively fast even on the ICR-3 (1.7 FPS), however, it has lower performance (28%). Therefore, the 256 x 256 is made available to provide more performance (45%) but slower execution speed (0.3 FPS on the ICR-3)
All in all, the Scailable library provides models for different tasks (locating, counting, raising alarms, etc.) with two versions to trade-off between speed and performance.
Please note that each one of these models has its detailed documentation. You can access it during the model configuration phase on the AI Manager.
Last modified 5d ago