Learning Cross-Video Neural Representations for High-Quality Frame Interpolation.