Fig. 4From: Domain randomization for neural network classificationOptimal accuracy of models trained using different synthetic datasets. “Synth” represents the control model trained on domain randomization techniques outlined in Sect. "Methods/experimental" using dropout. The others represent datasets were a certain parameter was omitted in the data generation processes such as lighting or textures. We also go a step further and texturize the cats and dogs in what is labelled as “texture subject” barBack to article page