This repo demonstrates how to use KotlinDL for neural network inference on Android devices. It contains a simple Android app that uses KotlinDL to demonstrate the inference of a bunch of pre-trained models for different computer vision tasks.

The list of demonstrated models includes:

• MobileNetV1 and EfficientNetV4Lite for image classification
• SSDMobileNetV1 and EfficientDetLite0 for object detection
• MoveNet for human pose estimation
• UltraFace320 for Face detection
• Fan2d106Face for Face Alignment

This application is based on CameraX Android API and uses the latest KotlinDL version. The actual model inference is performed by the Onnx Runtime.

This example is based on ort_image_classification example