How much real data do we actually need: Analyzing object detection performance using synthetic and real data
This addresses the costly data annotation problem in computer vision for researchers and practitioners, offering insights into efficient training methods.
The paper investigates how much real data is needed for object detection by analyzing performance when replacing real data with synthetic data and using limited real data, finding that synthetic data can effectively supplement or replace real data in training deep networks.
In recent years, deep learning models have resulted in a huge amount of progress in various areas, including computer vision. By nature, the supervised training of deep models requires a large amount of data to be available. This ideal case is usually not tractable as the data annotation is a tremendously exhausting and costly task to perform. An alternative is to use synthetic data. In this paper, we take a comprehensive look into the effects of replacing real data with synthetic data. We further analyze the effects of having a limited amount of real data. We use multiple synthetic and real datasets along with a simulation tool to create large amounts of cheaply annotated synthetic data. We analyze the domain similarity of each of these datasets. We provide insights about designing a methodological procedure for training deep networks using these datasets.