7月22日的Vision Detector应用程序分析

Vision Detector lets you run Core ML models directly on your iPhone or iPad without building an app in Xcode. It provides a simple way to test and evaluate machine learning models on real devices using on-device inference. To get started, create a machine learning model in Core ML format using Create ML or coremltools. Copy the model file to your device using the Files app. Supported locations include local storage and cloud services such as iCloud Drive, OneDrive, Google Drive, and Dropbox. Models can also be transferred using AirDrop. After launching Vision Detector, select and load your model. You can choose an input image from multiple sources: • Live video from the built-in camera with continuous inference • Still photos captured with the camera • Images from the Photo Library • Images from the file system For live video input, inference runs continuously on the camera feed. Performance such as frame rate depends on the device. Supported model types: • Image classification • Object detection • Style transfer Models that rely on MultiArray inputs or outputs, or that do not include a non-maximum suppression layer, are not supported. Vision Detector 3 can access large language modes (LLMs) using OpenAI compatible API Presets for OpenAI and Anthropic API are available with your own API key (iOS version only). IP address (192.168.1.24:1234) or machine host name (mac-mini.local:1234) are available. Default port number of LM Studio is 1234. You can customize user prompt. Most language model will follow in the language you use. Detailed output from LLMs spends more time to finish processing. Notification Center is available with prompt such as “Call notification tool in the case you see person”. Note: Vision Detector does not include any pre-trained models. You must provide your own CoreML model or access to API to use this app.