Research map / Video / Video segmentation and tracking
Video Frame Synthesis Using Deep Voxel Flow
Ziwei Liu, Raymond A. Yeh, Xiaoou Tang, Yiming Liu, Aseem Agarwala
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add multi-scale voxel flow to DVF (ablation) | Deep Voxel Flow · UCF-101 test set | PSNR | Improved |
| add skip connections to the network (ablation) | Deep Voxel Flow · UCF-101 | PSNR: 1.1dB | Improved |
| fine-tune on KITTI training set | Deep Voxel Flow · KITTI view synthesis | L1 error: 0.336 -> 0.178 | Improved |
| fine-tune on KITTI before evaluating flow | Deep Voxel Flow · KITTI flow 2012 | EPE: 14.6 -> 9.5 | Improved |
| fine-tune with action recognition loss after frame-synthesis pretraining | Deep Voxel Flow · UCF-101 action recognition | accuracy: 48.7 -> 52.4 | Improved |
| use DVF frame-synthesis pretraining instead of random initialization | action recognition network · UCF-101 action recognition | accuracy: 39.1 -> 52.4 | Improved |
| replace Beyond MSE hallucination with optical-flow pixel copying | video interpolation · UCF-101 interpolation | PSNR: 1.4dB | Improved |
| replace EpicFlow with learned voxel flow | video interpolation · UCF-101 interpolation | PSNR: 1.6dB | Improved |
| replace EpicFlow with learned voxel flow | video extrapolation · UCF-101 extrapolation (2 frames) | PSNR: 31.3 -> 32.7 | Improved |
| use 4 input frames instead of 2 for extrapolation | Deep Voxel Flow · UCF-101 extrapolation | PSNR: 32.7 -> 33.4 | Improved |
| apply UCF-101-trained DVF directly to KITTI view synthesis versus prior methods | Deep Voxel Flow · KITTI view synthesis | L1 error: 0.164 and 0.135 | Improved |
| use unsupervised DVF flow instead of EpicFlow for flow estimation | flow estimation · KITTI flow 2012 | EPE: 3.8 -> 9.5 | Hurt |
Closest papers by meaning
- Channel Attention Is All You Need for Video Frame Interpolation (AAAI 2020 · similarity 0.89)
- Context-Aware Synthesis for Video Frame Interpolation (CVPR 2018 · similarity 0.89)
- Depth-Aware Video Frame Interpolation (CVPR 2019 · similarity 0.88)
- Video Frame Interpolation via Adaptive Convolution (CVPR 2017 · similarity 0.88)
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.87)
- Hierarchical Flow Diffusion for Efficient Frame Interpolation (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
