Research map / Video / Video generation
StyleInV: A Temporal Style Modulated Inversion Network for Unconditional Video Generation
Yuhan Wang, Liming Jiang, 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 |
|---|---|---|---|
| replace autoregressive motion generator with inversion-based temporal style modulation (ablation) | StyleInV · DeeperForensics | FVD16: 59.49 (w/o inversion encoder) -> 41.58 | Improved |
| use first-frame-aware acyclic positional encoding (FFA-APE) (ablation) | StyleInV · DeeperForensics | FVD16: 88.98 -> 41.58 | Improved |
| add initial frame to discriminator and reconstruction loss (ablation) | StyleInV · DeeperForensics | FVD16: 67.43 -> 41.58 | Improved |
| add initial frame to discriminator while keeping L2 loss (ablation) | StyleInV · DeeperForensics | FVD16: 86.32 -> 41.58 | Improved |
| apply sparse training to MoCoGAN-HD baseline | MoCoGAN-HD · DeeperForensics | FVD128: 610.30 -> 129.13 | Improved |
| increase DIGAN clip length to 128 | DIGAN · DeeperForensics | FVD16: FVD16 46.69 -> 87.42; FVD128 1060.27 -> 95.80 | Mixed |
| inject constant noise in StyleGAN2 generator (ablation) | StyleInV · FaceForensics, TaiChi | FVD16: 106.42 -> 47.88 (FaceForensics) | Improved |
| inject constant noise in StyleGAN2 generator (ablation) | StyleInV · SkyTimelapse | FVD16: 77.04 -> 115.68 | Hurt |
| train parent model on FFHQ-cropped data for finetuning-based style transfer (ablation) | StyleInV · style transfer | qualitative structure | Improved |
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.
- StyleGAN-V — StyleGAN-V: A Continuous Video Generator With the Price, Image Quality and Perks of StyleGAN2 (CVPR 2022) · better on 4 of 7 shared results
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
- A Style-Based GAN Encoder for High Fidelity Reconstruction of Images and Videos (ECCV 2022 · similarity 0.92)
- StyleGAN-V: A Continuous Video Generator With the Price, Image Quality and Perks of StyleGAN2 (CVPR 2022 · similarity 0.91)
- MoStGAN-V: Video Generation With Temporal Motion Styles (CVPR 2023 · similarity 0.89)
- Temporal Generative Adversarial Nets With Singular Value Clipping (ICCV 2017 · similarity 0.88)
- Towards Smooth Video Composition (ICLR 2023 · similarity 0.88)
- StyleCineGAN: Landscape Cinemagraph Generation using a Pre-trained StyleGAN (CVPR 2024 · 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.
