Multi-task Data Augmentation Method Joint Object Detection and Semantic Segmentation

Data augmentation is a commonly used method for training networks in deep learning. For object detection tasks, data augmentation methods include random flip, random crop, etc. For semantic segmentation tasks, data enhancement includes horizontal flip, random saturation transformation, etc. With the application of deep learning technology in autonomous driving, many multi-task networks based on deep learning have proposed. These multi-task networks can simultaneously process target detection tasks and semantic segmentation tasks in autonomous driving. For the multi-task data augmentation, the existing methods can only take some simple methods. This paper proposed an effective data augmentation method, which can combine the data augmentation of object detection and semantic segmentation at the same time.