Research map / Probabilistic and scientific modeling / Molecules and proteins
LiteFlowNet3: Resolving Correspondence Ambiguity for More Accurate Optical Flow Estimation
Tak-Wai Hui, Chen Change Loy
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 cost volume modulation (without confidence map) (ablation) | LiteFlowNet3 · Sintel final / KITTI 2015 (train) | AEE: 4.14 -> 4.09; 11.31 -> 11.01 | Improved |
| add flow field deformation on top of modulation, no confidence map (ablation) | LiteFlowNet3 · Sintel final / KITTI 2012 (train) | AEE: 4.09 -> 4.06 (Sintel final); 4.02 -> 4.06 (KITTI 2012) | Mixed |
| add confidence map to cost volume modulation (ablation) | LiteFlowNet3 · Sintel final / KITTI 2015 (train) | AEE: 4.09 -> 4.02; 11.01 -> 10.65 | Improved |
| add flow field deformation to modulation with confidence map (ablation) | LiteFlowNet3 · Sintel final / KITTI 2015 (train) | AEE: 4.02 -> 3.91; 10.65 -> 10.40 | Improved |
| add all modules (modulation, deformation, confidence map) (ablation) | LiteFlowNet3 · Sintel clean / KITTI 2012 (train) | AEE: 2.78 -> 2.59; 4.11 -> 3.88 | Improved |
| augment cost volume with features of I1 and flow field instead of modulation (ablation) | LiteFlowNet3 · Sintel final / KITTI 2015 (train) | AEE: 4.14 -> 4.20; 11.12 -> 11.30 | Hurt |
| use cost volume modulation instead of feature/flow augmentation (ablation) | LiteFlowNet3 · KITTI 2012 (train) | AEE: 4.28 -> 3.95 | Improved |
| use cost volume modulation instead of no augmentation (ablation) | LiteFlowNet3 · Sintel final / KITTI 2015 (train, FlyingChairs) | AEE: 4.14 -> 4.02; 11.12 -> 10.65 | Improved |
| add weight sharing to decoders and modules (LiteFlowNet3-S) (ablation) | LiteFlowNet3 · model size | parameters (M): 7.5M -> 5.2M | Improved |
| use LiteFlowNet3 instead of LiteFlowNet2 | LiteFlowNet2 · Sintel clean (test) unmatched | AEE: 20.64 -> 18.08 | Improved |
| use LiteFlowNet3 instead of LiteFlowNet2 | LiteFlowNet2 · Sintel clean (test) | AEE: 3.48 -> 2.99 | Improved |
Closest papers by meaning
- LiteFlowNet: A Lightweight Convolutional Neural Network for Optical Flow Estimation (CVPR 2018 · similarity 0.93)
- A Lightweight Optical Flow CNN —Revisiting Data Fidelity and Regularization (TPAMI 2020 · similarity 0.93)
- Rethinking Optical Flow From Geometric Matching Consistent Perspective (CVPR 2023 · similarity 0.89)
- Learning Optical Flow From Still Images (CVPR 2021 · similarity 0.89)
- Unsupervised Deep Learning for Optical Flow Estimation (AAAI 2017 · similarity 0.88)
- FlowNet 2.0: Evolution of Optical Flow Estimation With Deep Networks (CVPR 2017 · similarity 0.88)
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.
