This repository contains an iOS application that uses the Nezha framework for real-time object detection. Nezha is a low-level cross-platform GPU computation framework that efficiently synchronizes machine learning (ML) and graphics GPU workloads using render graphs, providing high throughput and seamless interfacing with Vulkan, Metal, and CUDA. This project was developed as a part of Stanford's CS231N course.
The Nezha iOS App leverages Nezha's capabilities to perform real-time object detection using advanced GPU synchronization techniques and shader programming.
- Real-Time Object Detection: Utilizes the Nezha framework for efficient, real-time object detection on iOS devices.
- High-Performance Shaders: Implements shaders using GLSL to perform Convolutional Neural Network (CNN) layers at speeds comparable to PyTorch and TensorFlow.
- Cross-Platform Compatibility: Nezha seamlessly interfaces with Vulkan, Metal, and CUDA for broad compatibility.
- Objective-C Integration: The iOS app is developed using Objective-C for optimal performance.
- Launch the App: Open the Nezha iOS app on your device.
- Real-Time Object Detection: Use the app to perform real-time object detection. The app leverages the Nezha framework to efficiently process and detect objects in real-time.
Contributions are welcome! If you have suggestions or improvements, please create a pull request or open an issue on the GitHub repository.
This project builds upon the Nezha framework, leveraging its advanced GPU synchronization and shader programming capabilities for real-time object detection on iOS. Special thanks to the contributors and the open-source community for their support.
This project is licensed under the MIT License. See the LICENSE file for details.