Neural Object Detector 1.5

Neural Object Detector is a versatile application tailored for developers and enthusiasts in the fields of Machine Learning, Computer Vision, and Object Detection/Image Classification. It comes pre-packaged with the YOLOv3 model, a well-known neural network renowned for its speed and capability to detect 80 distinct classes of objects.

One of the standout features of Neural Object Detector is its ability to allow users to import custom machine learning models for object detection or image classification seamlessly. With just a tap, users can integrate downloaded models via the Files app import window, accessible by pressing the plus icon in the models view, eliminating the need to exit the application.

The application visually annotates detected objects by drawing rounded rectangles around them, and it provides options to render annotated images for saving or sharing, enhancing user experience. Additionally, it offers various configurable settings regarding computer vision algorithms and camera resolution, catering to diverse user preferences. Notably, users can enable CPU-only mode for testing models under specific conditions.

Beyond Object Detection, Neural also supports Image Classification through its bundled ResNet50 model. Currently, the application accepts Core ML models in the *.mlmodel format specifically for object detection; other model formats are not supported at this time.

The app is compatible with several Core ML supported tools, services, and converters, including:

• Turi Create
• IBM Watson Services
• Core ML Tools
• Apache MXNet
• TensorFlow
• ONNX

The YOLOv3 model included with Neural Object Detector is a free and open-source model. More information can be found on its GitHub repository. Similarly, the ResNet50 model for Image Classification is also free and open-source; details are available here.

For acknowledgments related to this application, users can visit this page.

Importantly, all machine learning processes occur on-device, ensuring that user data remain secure as no cloud services or private analytics services are utilized. Should users require support, minimal data will be compiled into a log file and attached to the email composer. This data encompasses app version, configuration, device model, battery information, software version, and CPU usage. Users have complete control over this data and may opt not to share it if they prefer.