AutoFlow: Learning a Better Training Set for Optical Flow

Model (𝜃) Target data Model (𝜃) Target data Typical pipeline Static FlyingChairs AutoFlow (𝜆) AutoFlow pipeline Figure 1: Left: Pipelines for optical flow. A typical pipeline pre-trains models on static datasets, e.g., FlyingChairs, and then evaluates the performance on a target dataset, e.g., Sintel. AutoFlow learns pre-training data which is optimized on a target dataset. Right: Accuracy w.r.t. number of pre-training examples on Sintel.final. Four AutoFlow pre-training examples with augmentation achieve lower errors than 22,872 FlyingChairs pre-training examples with augmentation. The gap between PWC-Net and RAFT becomes small when pre-trained on enough AutoFlow examples.