Learning optical flow from still images
This addresses the data scarcity problem for researchers and practitioners in computer vision, offering an incremental improvement by generating synthetic flow from real images to enhance training.
The paper tackles the scarcity of training data for optical flow networks by introducing a framework that generates accurate ground-truth optical flow annotations from single real images, using monocular depth estimation and virtual camera movements. When trained with this data, state-of-the-art optical flow networks achieve superior generalization to unseen real data compared to models trained on synthetic datasets or unlabeled videos, with better specialization when combined with synthetic images.
This paper deals with the scarcity of data for training optical flow networks, highlighting the limitations of existing sources such as labeled synthetic datasets or unlabeled real videos. Specifically, we introduce a framework to generate accurate ground-truth optical flow annotations quickly and in large amounts from any readily available single real picture. Given an image, we use an off-the-shelf monocular depth estimation network to build a plausible point cloud for the observed scene. Then, we virtually move the camera in the reconstructed environment with known motion vectors and rotation angles, allowing us to synthesize both a novel view and the corresponding optical flow field connecting each pixel in the input image to the one in the new frame. When trained with our data, state-of-the-art optical flow networks achieve superior generalization to unseen real data compared to the same models trained either on annotated synthetic datasets or unlabeled videos, and better specialization if combined with synthetic images.