Reply to the public account: course, to obtain resources.
RK3588 C++ Visual Deployment Course Introduction
1. Course Details
This course focuses on in-depth explanations of visual deployment using C++ on the RK3588 platform. As a leading domestic chip, the RK3588 features a quad-core Cortex-A76 and a quad-core Cortex-A55 processor architecture, combined with an integrated dedicated NPU unit, providing strong computational support for running deep learning models.
The course content starts with basic environment setup, guiding students to configure the Python environment on their computers, install essential tools such as YOLOv8 and ONNX, ensuring smooth model training and conversion; it then details how to configure the OpenCV environment on the RK3588, laying a solid foundation for subsequent model deployment. In the model training section, it provides downloads of the official YOLOv8 code and training steps, teaching students how to adjust configuration files based on custom datasets to enable accurate target recognition. In the model conversion part, it thoroughly introduces how to convert the trained PyTorch model to ONNX format, and further convert it to RKNN format, allowing the model to run efficiently on the RK3588 platform. Finally, it guides students to compile OpenCV on the RK3588 and deploy C++ inference code to achieve visual tasks such as video object detection.
2. Common Issues
During the environment setup phase, students often encounter various issues due to environment mismatches, such as incompatibilities with toolchains and code libraries like rknn-toolkit2 and rockchip-yolov8. The course will provide detailed troubleshooting and solutions for these common environmental issues, helping students successfully establish a stable development environment. During the model conversion process, errors may also occur, for example, when converting their own .pt model to ONNX, it may fail due to not modifying the model category information in yolov8s.yaml or selecting the wrong yaml file, leading to errors. The course will analyze these model conversion issues in depth with practical cases, enabling students to master problem-solving approaches and techniques.
3. How to Learn
To learn this course, it is recommended that students have a certain foundation in C++ programming and computer vision theory, which will help them better understand the course content. During the learning process, students should keep pace with the course rhythm, following the course steps to practice hands-on, from environment setup, model training to final deployment, completing each step themselves to deepen their understanding and mastery of the knowledge. When encountering problems, they should fully utilize the Q&A resources provided by the course, actively communicate and discuss with the instructor and other students to solve problems together. Additionally, they should be good at summarizing experiences, recording the problems and solutions encountered in each practice to form their own knowledge accumulation.
4. Course Value
Through this course, students will master the core skills of visual deployment using C++ on the RK3588 platform, enabling them to independently implement simple applications such as object detection and image recognition. This skill has a wide range of applications in popular fields such as security monitoring, autonomous driving assistance, and industrial inspection. For example, in the security monitoring field, it can achieve real-time recognition and tracking of suspicious targets, enhancing the intelligence level of monitoring systems; in autonomous driving, it can help vehicles recognize pedestrians, vehicles, and other targets on the road in real-time, ensuring driving safety. After mastering this technology, students will have a stronger competitive edge in the job market and related project practices, laying a solid foundation for future career development and technical research.
Are you particularly interested in a certain part of the course, or do you want to know more about the course practice cases? Feel free to let me know, and I can provide you with more detailed content.