ML Trainer: Make Training Data 1.0

The process of generating original training data for Image Classification machine learning models has been significantly streamlined with this tool.

ML Trainer enables developers to efficiently capture and export thousands of images directly to the Photos application, facilitating seamless import through iCloud or the built-in Image Capture utility on macOS.

Users can initiate image collection by pressing the Scan button, which captures a predetermined number of images as they move closer to or rotate around the subject. Alternatively, tapping the Camera button allows for the capture of a single image. The Flashlight toggle feature assists in improving image quality in low-light environments, while the Save button exports all captured images to the Photos app.

Images are consistently captured at a rate of 3 frames per second; however, the Frame Count for each scan can be customized within the Settings menu. Adjusting this parameter balances time efficiency against accuracy—higher Frame Counts reduce total capture time but may compromise detail across different angles, whereas lower counts enhance precision. Additionally, enabling Crosshairs and Guides in Settings can further improve image consistency.

This application is specifically optimized to accelerate data acquisition for integration with CreateML within Xcode. The exported images are compatible with other machine learning frameworks such as TensorFlow and Azure Machine Learning. For optimal data transfer speed, connecting your device via a wired connection to a macOS computer with the Image Capture app active is recommended.