Research map / Video / Video segmentation and tracking
Deep 360° Optical Flow Estimation Based on Multi-Projection Fusion
Yiheng Li, Connelly Barnes, Kun Huang, Fang-Lue Zhang
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 |
|---|---|---|---|
| fuse equirectangular and tri-cylindrical predictions (ablation) | single equirectangular projection · City100+EFT100 average | SD/EPE: 1.94/4.91 -> 1.74/3.40 | Improved |
| fuse cube-padding and tri-cylindrical predictions without equirectangular (ablation) | single equirectangular projection · City100+EFT100 average | EPE: 4.91 -> 3.60 | Improved |
| fuse equirectangular and cube-padding predictions (ablation) | equirectangular projection · City100+EFT100 average | EPE: 4.91 -> 3.89 | Improved |
| use tri-cylindrical rather than cube-padding as the second fused projection (ablation) | E+P fusion · City100+EFT100 average | EPE: 3.89 -> 3.40 | Improved |
| finetune with cube-padding instead of equirectangular (ablation) | PWC-Net · City100+EFT100 average | EPE: 4.91 -> 5.68 | Hurt |
| replace regular kernels with spherical kernels (ablation) | PWC-Net · City100UR | EPE: 3.17 -> 3.80 | Hurt |
| use RAFT instead of PWC-Net as finetuned backbone (ablation) | finetuned optical flow baseline · City100UR | EPE: 3.17 -> 4.72 | Hurt |
| add latitude and longitude channels to the fusion network input (ablation) | fusion network · City100+EFT100 | EPE: 2.61 -> 2.65 | Hurt |
| apply multi-projection fusion in place of equirectangular PWC-Net | PWC-Net · City100+EFT100 average | EPE: 12.74 -> 3.40 | Improved |
| use fusion approach instead of tangent image method | TanImg · City100+EFT100 average | EPE: 5.88 -> 3.40 | Improved |
| add color data augmentation (RGBShift, ChannelShuffle) on small circle set | fine-tuned model · TanImg circle dataset | EPE: 4.73 -> 3.8 | Improved |
Closest papers by meaning
- PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View (ICCV 2025 · similarity 0.86)
- Learning Optical Flow From Still Images (CVPR 2021 · similarity 0.85)
- Unsupervised Deep Learning for Optical Flow Estimation (AAAI 2017 · similarity 0.85)
- A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation (CVPR 2016 · similarity 0.85)
- Feature-Level Collaboration: Joint Unsupervised Learning of Optical Flow, Stereo Depth and Camera Motion (CVPR 2021 · similarity 0.85)
- Offline and Online Optical Flow Enhancement for Deep Video Compression (AAAI 2024 · similarity 0.84)
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.
