Research map / Video / Video segmentation and tracking
Self-Supervised Video GANs: Learning for Appearance Consistency and Motion Coherency
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 appearance contrastive learning and temporal structure puzzle to MoCoGAN | MoCoGAN · Weizmann, MUG, UvA-NEMO, and UCF101 | FVD: 194.34 -> 189.65; 102.20 -> 90.79; 46.20 -> 40.56; 869.41 -> 822.48 | Helped |
| Add appearance contrastive learning and temporal structure puzzle to G3AN | G3AN · Weizmann, MUG, UvA-NEMO, and UCF101 | FVD: 117.69 -> 105.51; 89.73 -> 67.62; 56.97 -> 39.62; 687.67 -> 643.55 | Helped |
| Add both proposed self-supervision losses rather than using the MoCoGAN baseline | MoCoGAN · UCF101 | IS: 1.46 -> 1.48 | Helped |
| Add appearance contrastive loss (ablation) | MoCoGAN · UvA-NEMO and UCF101 | FVD: 46.20 -> 41.56; 869.41 -> 853.89 | Helped |
| Add temporal structure loss (ablation) | MoCoGAN · UvA-NEMO and UCF101 | FVD: 46.20 -> 43.40; 869.41 -> 856.76 | Helped |
| Use both appearance contrastive and temporal structure losses instead of either loss alone (ablation) | MoCoGAN · UvA-NEMO and UCF101 | FVD: 41.56 -> 40.56; 43.40 -> 40.56; 853.89 -> 822.48; 856.76 -> 822.48 | Helped |
| Replace appearance contrastive learning with rotation prediction (ablation) | MoCoGAN · UvA-NEMO and UCF101 | FVD: 40.56 -> 46.05; 822.48 -> 859.30 | Hurt |
| Replace appearance contrastive learning with rotation prediction (ablation) | MoCoGAN · UCF101 | IS: 1.48 -> 1.38 | Hurt |
| Remove temporal sampling from appearance contrastive learning (ablation) | MoCoGAN · UvA-NEMO | FVD: 41.56 -> 42.54 | Hurt |
| Remove temporal sampling from appearance contrastive learning (ablation) | MoCoGAN · UCF101 | FVD: 853.89 -> 921.73 | Hurt |
| Use video-level rather than clip-level input for temporal structure puzzle (ablation) | MoCoGAN · UvA-NEMO and UCF101 | FVD: 55.44 -> 43.40; 1004.68 -> 856.76 | Helped |
Closest papers by meaning
- Video Prediction with Appearance and Motion Conditions (ICML 2018 · similarity 0.90)
- Hierarchical Video Generation From Orthogonal Information: Optical Flow and Texture (AAAI 2018 · similarity 0.89)
- G3AN: Disentangling Appearance and Motion for Video Generation (CVPR 2020 · similarity 0.88)
- Generating Videos with Scene Dynamics (NeurIPS 2016 · similarity 0.88)
- VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models (CVPR 2025 · similarity 0.85)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.85)
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.