Here’s an introduction to the “MobileObjectLocalizer,” a machine learning model available for use with the ailia SDK. By utilizing this model, available on ailia MODELS, you can easily implement AI functionalities into applications.

About MobileObjectLocalizer

MobileObjectLocalizer is a general-purpose object detection model developed by Google. Unlike the typical YOLO model that recognizes 80 classes from COCO, MobileObjectLocalizer is not limited to predefined classes and can detect any object.

Architecture of MobileObjectLocalizer

MobileObjectLocalizer uses MobileNetV2 and SSD-Lite architectures. It accepts an input resolution of 192×192 and can output up to 100 bounding boxes. While its accuracy is not the highest, it can detect bounding boxes for arbitrary objects without the need for training. This allows for various applications, such as incorporating a ResNet50 to create a 1000-class classifier or using it for autofocus by detecting the bounding box with the highest confidence value displayed on the screen.

How to Use MobileObjectLocalizer with ailia SDK

To use MobileObjectLocalizer in ailia SDK, the following command is used:

$ python3 mobile_object_localizer.py -v 0

Ax Inc., the company behind ailia SDK, specializes in practical AI applications. They offer cross-platform, GPU-accelerated inference capabilities and provide a complete solution for AI that includes consulting, model creation, SDK provision, development of AI-driven apps and systems, and support. Feel free to contact them for more information on their services.