FlowNet 2.0 improves optical flow estimation by enhancing end-to-end learning. Key contributions include optimizing training data schedules, introducing a stacked architecture with image warping, and developing a subnetwork for small displacements. FlowNet 2.0 achieves significant accuracy improvements and maintains interactive frame rates. It performs on par with state-of-the-art methods and runs at up to 140fps. The network is trained on diverse datasets, including Chairs and Things3D, and includes a specialized dataset for small displacements. Stacking networks with warping improves results, while a fusion network combines estimates for better performance. FlowNet 2.0 outperforms previous versions on benchmarks like Sintel and KITTI, and excels in real-world applications. The architecture is efficient, scalable, and robust to motion blur and compression artifacts. FlowNet 2.0 demonstrates strong performance in motion segmentation and action recognition, showing its effectiveness in various scenarios. The paper highlights the importance of training data schedules and network stacking for achieving high accuracy and speed in optical flow estimation.FlowNet 2.0 improves optical flow estimation by enhancing end-to-end learning. Key contributions include optimizing training data schedules, introducing a stacked architecture with image warping, and developing a subnetwork for small displacements. FlowNet 2.0 achieves significant accuracy improvements and maintains interactive frame rates. It performs on par with state-of-the-art methods and runs at up to 140fps. The network is trained on diverse datasets, including Chairs and Things3D, and includes a specialized dataset for small displacements. Stacking networks with warping improves results, while a fusion network combines estimates for better performance. FlowNet 2.0 outperforms previous versions on benchmarks like Sintel and KITTI, and excels in real-world applications. The architecture is efficient, scalable, and robust to motion blur and compression artifacts. FlowNet 2.0 demonstrates strong performance in motion segmentation and action recognition, showing its effectiveness in various scenarios. The paper highlights the importance of training data schedules and network stacking for achieving high accuracy and speed in optical flow estimation.