Challenges and limitations
In this section, we will highlight the main challenges in using this approach for synthetic data generation. We will look at realism, diversity, and complexity issues that present some difficulties in utilizing this approach for synthetic data generation.
Realism
The domain gap between synthetic and real data is one of the main issues that limit the usability of synthetic data. For synthetic data to be useful, it should mimic the distribution and statistical characteristics of its real counterparts. Thus, for computer vision problems, we need to ensure a high degree of photorealism, otherwise, ML models trained on synthetic data may not generalize well to real data.
Achieving a high degree of photorealism using game engines and simulators is not a simple task. Even with the help of contemporary game engines such as CryEngine, Unreal, and Unity, we need effort, skill, and time to create photorealistic scenes.
The three essential elements for approaching...
