Research map / Video / Video segmentation and tracking
Learning Temporally and Semantically Consistent Unpaired Video-to-Video Translation through Pseudo-Supervision from Synthetic Optical Flow
Kaihong Wang, Kumar Akash, Teruhisa Misu
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 |
|---|---|---|---|
| Replace estimated motion guidance with consistent synthetic optical flow | Unpaired video-to-video translation · Viper-to-Cityscapes | warping error: 0.039703 -> 0.035984 | Improved |
| Add unsupervised recycle and spatial losses together (ablation) | Video-to-labels translation · Viper | mIoU: 0.16 -> 9.89 | Improved |
| Add unsupervised recycle loss to the spatial loss (ablation) | Video-to-labels translation · Viper | mIoU: 0.16 -> 9.89 | Improved |
| Add video domain regularization suppression (ablation) | UR + US video-to-labels model · Viper labels-to-video | MP: 29.50 -> 37.93 | Improved |
| Add cycle loss (ablation) | Video-to-labels model with UR, US, and VDRS · Viper video-to-labels | mIoU: 9.53 -> 12.29 | Improved |
| Add cycle loss (ablation) | Viper-to-Cityscapes translation with UR and US · Viper-to-Cityscapes | warping error: 0.049261 -> 0.047391 | Improved |
| Add content loss (ablation) | Viper-to-Cityscapes translation with UR and US · Viper-to-Cityscapes | warping error: 0.049261 -> 0.035984 | Improved |
| Replace synthetic flow with estimated optical flow (ablation) | Proposed translation method · Viper-to-Cityscapes | warping error: 0.035984 -> 0.048983 | Hurt |
| Use translation-only synthetic flow (ablation) | Proposed translation method · Viper-to-Cityscapes | warping error: 0.035984 -> 0.050842 | Hurt |
| Use scaling-only synthetic flow (ablation) | Proposed translation method · Viper-to-Cityscapes | warping error: 0.035984 -> 0.037798 | Hurt |
| Remove Gaussian noise from synthetic future-frame generation (ablation) | Proposed translation method · Viper video-to-labels | mIoU: 12.29 -> 0.16 | Hurt |
| Use mismatched synthetic flows across source and target domains (ablation) | Proposed translation method · Viper video-to-labels | mIoU: 12.29 -> 0.46 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Recycle-GAN — Recycle-GAN: Unsupervised Video Retargeting (ECCV 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.87)
- Optical-Flow Guided Prompt Optimization for Coherent Video Generation (CVPR 2025 · similarity 0.86)
- MeDM: Mediating Image Diffusion Models for Video-to-Video Translation with Temporal Correspondence Guidance (AAAI 2024 · similarity 0.86)
- Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models (AAAI 2025 · similarity 0.86)
- MotionRAG: Motion Retrieval-Augmented Image-to-Video Generation (NeurIPS 2025 · similarity 0.86)
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.
