車載カメラを用いた物体検出における実環境とCG環境の一致性検証
- Author
- 山下隆義, 板谷英典, 平川翼, 藤吉弘亘, 長瀬功児, 小山翔太郎, 井上秀雄
- Publication
- 自動車技術会春季大会, 2023
Download: PDF (Japanese)
Training and evaluating of object detection models require a large amount of diverse data, but it is impractical to create such data in a real environment. Therefore, it is expected that data can be created using a computer graphics environment. However, there is a problem that there are differences in the domain between the real and CG environments. Therefore, the CG environment is required to have the same reproducibility as the real environment. In our research, we clarify the differences between real and CG environments based on the detection accuracy of models in the same scene, and examine the problems of CG environments. In addition, we will examine scenes in which models are verified using the CG environment.