This repository provides a TensorFlow implementation of FlowNet 2.0, a deep learning architecture for optical flow estimation in computer vision. It includes multiple model variants (FlowNetC, S, CS, CSS, SD, and the stacked FlowNet2) and supports both inference and training. The models use convolutional neural networks to estimate motion vectors between consecutive image frames. The repository includes pre-trained checkpoints, benchmark timing data, and scripts for generating flow predictions from image pairs.