Temporal Interpolation as an Unsupervised Pretraining Task for Optical Flow Estimation
This approach addresses the data annotation bottleneck for low-level video tasks like optical flow, offering a practical solution for real-world applications, though it is incremental in combining unsupervised pretraining with fine-tuning.
The paper tackles the challenge of training CNNs for optical flow estimation without extensive annotated data by using unsupervised temporal interpolation as a pretraining task, and after fine-tuning on small ground truth data, it outperforms supervised methods using synthetic flow.
The difficulty of annotating training data is a major obstacle to using CNNs for low-level tasks in video. Synthetic data often does not generalize to real videos, while unsupervised methods require heuristic losses. Proxy tasks can overcome these issues, and start by training a network for a task for which annotation is easier or which can be trained unsupervised. The trained network is then fine-tuned for the original task using small amounts of ground truth data. Here, we investigate frame interpolation as a proxy task for optical flow. Using real movies, we train a CNN unsupervised for temporal interpolation. Such a network implicitly estimates motion, but cannot handle untextured regions. By fine-tuning on small amounts of ground truth flow, the network can learn to fill in homogeneous regions and compute full optical flow fields. Using this unsupervised pre-training, our network outperforms similar architectures that were trained supervised using synthetic optical flow.