A brief review of different architectures of Optical Flow algorithms

Courtesy: HackerEarth

Introduction

Ways to estimate optical flow

CNN regression architectures

1. FlowNet

Figure1: FlowNetSimple & FlowNetCorr architectures

2. Flownet 2.0

Figure2: FlowNet 2.0 architecture

3. SPyNet

Figure3: SPyNet architecture

4. LiteFlowNet

Figure4: LiteFlowNet architecture

5. PWCNet

Figure5: PWCNet

6. IRR-PWCNet

Figure6: FlowNet 2.0: stack of networks
Figure7: Iterative Residual Refinement (IRR) module
Figure8: A rolled version of Figure7
Figure9: Joint flow and occlusion estimation
Figure10: IRR-PWCNet architecture

Important points

References

Machine Learning Enthusiast