A Fine-Tuning Approach for Object Detection Models Using Virtually Generated Data

In this paper, we propose a fine-tuning approach to enhance the detection models using generated data in virtual environments. At first, we investigated the details of object detection models and analyzed their structures. For training models, we generated data within a virtual environment constructed using the Isaac Sim module on the NVIDIA Omniverse platform. We evaluated the performance of our trained models on both virtual environment data and real-world data, and tuned our models with real data given the performance results. Our final model results showed the possible enhancement of their object detection capabilities with virtual and real data.