Learn2Synth: Learning Optimal Data Synthesis Using Hypergradients

Domain randomization through synthesis is a powerful strategy to train networks that are unbiased with respect to the domain of the input images. Randomization allows networks to see a virtually infinite range of intensities and artifacts during training, thereby minimizing overfitting to appearance and maximizing generalization to unseen data. While powerful, this approach relies on the accurate tuning of a large set of hyper-parameters governing the probabilistic distribution of the synthesized images. Instead of manually tuning these parameters, we introduce Learn2Synth, a novel procedure in which synthesis parameters are learned using a small set of real labeled data. Unlike methods that impose constraints to align synthetic data with real data (e.g., contrastive or adversarial techniques), which risk misaligning the image and its label map, we tune an augmentation engine such that a segmentation network trained on synthetic data has optimal accuracy when applied to real data. This approach allows the training procedure to benefit from real labeled examples, without ever using these real examples to train the segmentation network, which avoids biasing the network towards the properties of the training set. Specifically, we develop both parametric and nonparametric strategies to augment the synthetic images, enhancing the segmentation network's performance. Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of this learning strategy. Code is available at: https://github.com/HuXiaoling/Learn2Synth.