Multigrid Predictive Filter Flow for Unsupervised Learning on Videos
About
We introduce multigrid Predictive Filter Flow (mgPFF), a framework for unsupervised learning on videos. The mgPFF takes as input a pair of frames and outputs per-pixel filters to warp one frame to the other. Compared to optical flow used for warping frames, mgPFF is more powerful in modeling sub-pixel movement and dealing with corruption (e.g., motion blur). We develop a multigrid coarse-to-fine modeling strategy that avoids the requirement of learning large filters to capture large displacement. This allows us to train an extremely compact model (4.6MB) which operates in a progressive way over multiple resolutions with shared weights. We train mgPFF on unsupervised, free-form videos and show that mgPFF is able to not only estimate long-range flow for frame reconstruction and detect video shot transitions, but also readily amendable for video object segmentation and pose tracking, where it substantially outperforms the published state-of-the-art without bells and whistles. Moreover, owing to mgPFF's nature of per-pixel filter prediction, we have the unique opportunity to visualize how each pixel is evolving during solving these tasks, thus gaining better interpretability.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Video Object Segmentation | DAVIS 2017 (val) | J mean42.2 | 1130 | |
| One-shot Video Object Segmentation | DAVIS 2016 (val) | J Mean40.5 | 28 | |
| Instance Segmentation Propagation | DAVIS 2017 | J Mean42.2 | 14 | |
| Human Pose Tracking | JHMDB (split1) | PCK @ 0.158.4 | 11 | |
| One-shot Video Object Segmentation | DAVIS 2017 (val) | J&F Mean44.6 | 11 | |
| Video Frame Reconstruction | DAVIS 2017 (val) | Pixel L1 Dist7.32 | 8 |