Research map / Video / Video generation
Pose Guided Human Video Generation
Ceyuan Yang, Zhe Wang, Xinge Zhu, Chen Huang, Jianping Shi, Dahua Lin
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 adversarial loss instead of L1-only training (ablation) | PSGAN · Human Action Dataset (UCF101 + Weizmann) | average pairwise L2 distance (lower is better): 0.0124 -> 0.0064 | Improved |
| add adversarial loss instead of L1-only training (ablation) | PSGAN · Human Facial Dataset (CK+) | average pairwise L2 distance (lower is better): 0.0078 -> 0.0051 | Improved |
| add LSTM/GRU temporal module (ablation) | PSGAN · Human Action Dataset (UCF101 + Weizmann) | average pairwise L2 distance (lower is better): 0.0072 -> 0.0064 | Improved |
| add LSTM/GRU temporal module (ablation) | PSGAN · Human Facial Dataset (CK+) | average pairwise L2 distance (lower is better): 0.0062 -> 0.0051 | Improved |
| add adversarial loss instead of L1-only training (ablation) | PSGAN · Human Action Dataset (UCF101 + Weizmann) user study | fraction ranking 1st: 0.09 -> 0.32 | Improved |
| add LSTM/GRU temporal module (ablation) | PSGAN · Human Action Dataset (UCF101 + Weizmann) user study | fraction ranking 1st: 0.21 -> 0.32 | Improved |
| add predicted-pose stream with semantic consistency to ground-truth-pose training (SCGAN vs SCGAN-gen) (ablation) | SCGAN · Human Action Dataset (UCF101 + Weizmann) | SSIM (higher is better): 0.73 -> 0.87 | Improved |
| add predicted-pose stream with semantic consistency to ground-truth-pose training (SCGAN vs SCGAN-gen) (ablation) | SCGAN · Human Action Dataset (UCF101 + Weizmann) | LPIPS (lower is better): 0.083 -> 0.041 | Improved |
| add predicted-pose stream with semantic consistency to ground-truth-pose training (SCGAN vs SCGAN-gen) (ablation) | SCGAN · Human Facial Dataset (CK+) | SSIM (higher is better): 0.89 -> 0.91 | Improved |
| add predicted-pose stream with semantic consistency to ground-truth-pose training (SCGAN vs SCGAN-gen) (ablation) | SCGAN · Human Facial Dataset (CK+) | LPIPS (lower is better): 0.038 -> 0.026 | Improved |
| train on predicted plus ground-truth poses (SCGAN) vs ground-truth only (SCGAN-gt) (ablation) | SCGAN · Human Action Dataset (UCF101 + Weizmann) | SSIM (higher is better): 0.89 -> 0.87 | No clear effect |
| use pose-guided PSGAN+SCGAN instead of MoCoGAN | Ours · Human Action Dataset (UCF101 + Weizmann) | Inception Score: 4.02 -> 5.70 | 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.
- MoCoGAN — MoCoGAN: Decomposing Motion and Content for Video Generation (CVPR 2018) · better on 2 of 2 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
- Synthesizing Images of Humans in Unseen Poses (CVPR 2018 · similarity 0.90)
- Scene-Aware Generative Network for Human Motion Synthesis (CVPR 2021 · similarity 0.89)
- The Pose Knows: Video Forecasting by Generating Pose Futures (ICCV 2017 · similarity 0.88)
- Pose Guided Person Image Generation (NeurIPS 2017 · similarity 0.88)
- Deep Video Generation, Prediction and Completion of Human Action Sequences (ECCV 2018 · similarity 0.88)
- Human Motion Transfer With 3D Constraints and Detail Enhancement (TPAMI 2022 · 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.
