Imagine building a vision-based object detection application for the production floor of a factory. Traditional methods take a lot of time to collect data samples, and when the layout of the factory floor changes, you need to re-collect a lot of data. This process is time-consuming, labor-intensive, and places high demands on the quality and accuracy of the data.
Now we can supplement the training dataset with data synthesized in the virtual world using synthetic data technology, which is also the theme of NVIDIA's 8th Sky Hackathon competition.
Co-hosted by the NVIDIA Enterprise Developer Community and Tronec, the three-week competition attracted 50 teams from 48 universities and 245 participants (including mentors). After intense competition, 29 teams successfully completed the competition.
With the theme of "Carton Inspection on Conveyor Belts on the Factory Floor", the participating teams experienced a series of challenges and learning opportunities during the competition. The activities covered data synthesis, model training, model inference, and UI page construction.
During the data synthesis phase, each team uses Omniverse Replicator for data synthesis based on the 3D models and scenes provided. This process is designed to help students become familiar with the process of synthesizing, collecting, filtering, cleaning, and labeling deep learning datasets.
Next, the participating teams need to set up an environment on their own servers according to the tutorials provided by the organizing committee, and train the target detection model. This step allows students to experience first-hand the training process of a deep learning model based on object recognition.
After the training is completed, the participating teams will deploy the trained model on the Jetson Xavier NX cluster provided by the organizing committee, and use NVIDIA TensorRT to accelerate the inference process of the model. This process taught students the importance of using NVIDIA TAO to optimize models such as pruning, as well as the ability to deploy TensorRT.
In order to provide a better user interaction experience, each team also needs to build its own UI page to create a complete application system.
By participating in this event, participants can not only learn the actual process of deep learning model training, but also experience the knowledge of Omniverse Replicator synthetic training data, NVIDIA TensorRT model deployment, and Jetson Xavier NX application in real scenarios.
In the end, students from Beijing University of Civil Engineering and Architecture, Zhejiang University Ningbo Institute of Technology, Southeast University and University of Science and Technology of China won the top three and received the NVIDIA Jetson Orin NANO development kit sponsored by Tronom Corporation. Fourth to 20th place will also receive the Jetson NANO development kit sponsored by Tronelb.
"NVIDIA Omniverse is a good supporter of data-centric deep learning research practices; The NVIDIA Tao Development Kit makes model training and deployment that requires high engineering requirements very convenient. Sky hackathon makes sense for the mass adoption of AI! Liu Mengyang, the captain of the team from Southeast University and the third winning team of this competition, said.
"This competition aims to open the door to the field of artificial intelligence for the participants, laying a solid foundation for their future learning and career development. Through hands-on operation and teamwork, the participants not only improved their technical level, but also cultivated the ability to solve problems and innovate. NVIDIA GPU application marketing director Hou Yutao said.
"We look forward to future competitions and events that will provide opportunities for more students to showcase their talents and creativity." At the same time, we also hope that such competitions can further promote the development of artificial intelligence technology and bring more innovation and progress to industry and society. Pu Mao, general manager of Chuanglebo, said.
Thank you to all participants and supporters for their efforts and support in making this event a complete success. Let's look forward to the future development and breakthrough in the field of artificial intelligence!