Research map / Video / Action recognition
Convolutional Sequence Generation for Skeleton-Based Action Synthesis
Sijie Yan, Zhizhong Li, Yuanjun Xiong, Huahan Yan, 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 |
|---|---|---|---|
| use a spectrum of Gaussian process length scales (0~100) instead of independent noise (length scale 0) (ablation) | CSGN · NTU-RGB+D | FID ensemble: 12.60 -> 5.86 | Improved |
| use a spectrum of Gaussian process length scales (0~100) instead of independent noise (length scale 0) (ablation) | CSGN · MikuDance | FID ensemble: 1016.87 -> 23.49 | Improved |
| use a spectrum of length scales (0~100) instead of constant latents (infinite length scale) (ablation) | CSGN · MikuDance | FID ensemble: 652.48 -> 23.49 | Improved |
| use a spectrum of length scales (0~100) instead of the best single length scale (ablation) | CSGN · NTU-RGB+D | FID ensemble: 11.77 -> 5.86 | Improved |
| use a spectrum of length scales (0~100) instead of the best single length scale (ablation) | CSGN · MikuDance | FID ensemble: 31.97 -> 23.49 | Improved |
| use a single large length scale (100) (ablation) | CSGN · MikuDance | FID ensemble: 31.97 -> 1552.46 | Hurt |
| widen spectrum range from 0~100 to 0~1000 (ablation) | CSGN · NTU-RGB+D | FID ensemble: 5.86 -> 6.19 | No clear effect |
| add graph convolution to temporal convolution (Temp-Conv to ST-GCN) (ablation) | Temp-Conv · NTU-RGB+D | FID ensemble: 22.25 -> 8.72 | Improved |
| add graph upsampling to ST-SGCN (full CSGN) (ablation) | ST-SGCN · NTU-RGB+D | FID ensemble: 101.2 -> 5.86 | Improved |
| add spatially varying graph convolution to CSGN-GC (graph conv plus upsampling) (ablation) | CSGN-GC · NTU-RGB+D | FID ensemble: 31.51 -> 5.86 | Improved |
| add spatially varying graph convolution to CSGN-GC (ablation) | CSGN-GC · MikuDance | FID ensemble: 31.20 -> 23.49 | Improved |
| add spatially varying graph convolution without upsampling (ST-GCN to ST-SGCN) (ablation) | ST-GCN · NTU-RGB+D | FID ensemble: 8.72 -> 101.2 | Hurt |
Closest papers by meaning
- Learning Diverse Stochastic Human-Action Generators by Learning Smooth Latent Transitions (AAAI 2020 · similarity 0.90)
- Structure-Aware Human-Action Generation (ECCV 2020 · similarity 0.90)
- Graph CNNs with Motif and Variable Temporal Block for Skeleton-Based Action Recognition (AAAI 2019 · similarity 0.90)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.90)
- Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition (AAAI 2018 · similarity 0.89)
- LAC - Latent Action Composition for Skeleton-based Action Segmentation (ICCV 2023 · similarity 0.89)
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.
