Conditional Patch-Based Domain Randomization: Improving Texture Domain Randomization Using Natural Image Patches

Using Domain Randomized synthetic data for training deep learning systems is a promising approach for addressing the data and the labeling requirements for supervised techniques to bridge the gap between simulation and the real world. We propose a novel approach for generating and applying class-specific Domain Randomization textures by using randomly cropped image patches from real-world data. In evaluation against the current Domain Randomization texture application techniques, our approach outperforms the highest performing technique by 4. 94 AP and 6. 71 AP when solving object detection and semantic segmentation tasks on the YCB-M [1] real-world robotics dataset. Our approach is a fast and inexpensive way of generating Domain Randomized textures while avoiding the need to handcraft texture distributions currently being used.

Authors

• Mohammad Ani
• Hector Basevi
• Ales Leonardis

Keywords

• Training
• Deep learning
• Semantic segmentation
• Object detection
• Data models
• Labeling
• Task analysis
• Natural Images
• Image Patches
• Domain Adaptation
• Natural Image Patches
• Real-world Data
• Detection Task
• Highest Performance
• Real-world Datasets
• Segmentation Task
• Object Detection Task
• Semantic Segmentation Task
• Training Set
• Understanding Of Function
• Type Of Approach
• Bounding Box
• Generative Adversarial Networks
• Object Classification
• RGB Images
• Real-world Images
• Object Dataset
• Target Dataset
• Texture Of Objects
• Checkers
• Robotic Tasks
• Zigzag Pattern
• Synthetic Images
• Household Objects
• Real-world Test