Closed Loop Dynamic Driving Data Mixture for Real-Synthetic Co-Training
Researchers have developed AutoScale, a novel closed-loop data engine designed to optimize the mixture of real and synthetic data for training autonomous driving models. This system dynamically adjusts the data composition based on performance feedback, addressing challenges like distribution shifts and inefficient data usage. AutoScale utilizes Graph Regularized AutoEncoder for scene representation and Cluster-aware Gradient Ascent for sample reweighting, demonstrating improved performance with fewer synthetic samples in experiments. AI
IMPACT This research could lead to more efficient training of autonomous driving systems by optimizing data mixtures.