Research map / Video / Action recognition
Synthetic Humans for Action Recognition from Unseen Viewpoints
Gül Varol, Ivan Laptev, Cordelia Schmid, Andrew Zisserman
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 synthetic videos to real training | ResNeXt-101 3D RGB · NTU CVS unseen 90 deg view | accuracy (%): 53.6 -> 69.0 | Improved |
| add synthetic videos to real training (UESTC Cross View I RGB) | ResNeXt-101 3D RGB · UESTC Cross View I | accuracy (%): 49.4 -> 66.4 | Improved |
| add synthetic videos to real training (RGB+Flow) | ResNeXt-101 3D · UESTC Cross View II | accuracy (%): 81.7 -> 87.4 | Improved |
| add synthetic videos to real training | ResNeXt-101 3D RGB+Flow · NTU CS | accuracy (%): 90.0 -> 90.7 | Improved |
| add optical flow stream (late fusion) (ablation) | Synth+Real RGB · NTU CVS 90 deg | accuracy (%): 69.0 -> 71.1 | Improved |
| use VIBE instead of HMMR motion estimates for rendering (ablation) | Synthetic data · NTU CVS 90 deg, RGB synthetic only | accuracy (%): 42.7 -> 45.3 | Improved |
| add more views to synthetic training (all 8 vs 0 deg only) (ablation) | Synthetic-only training · NTU CVS 90 deg | accuracy (%): 17.9 -> 38.8 | Improved |
| add video-level additive noise on joint rotations (ablation) | Synthetic training, 10 seq/action · NTU CVS 0 deg | accuracy (%): 33.1 -> 37.9 | Improved |
| add frame-level additive noise on joint rotations (ablation) | Synthetic training · NTU CVS 0 deg | accuracy (%): 33.1 -> 25.0 | Hurt |
| add pose interpolation augmentation (ablation) | Synthetic training · NTU CVS 0 deg | accuracy (%): 33.1 -> 35.7 | Mixed |
| add multi-person videos with xy translation (ablation) | Synthetic training · NTU CVS 0 deg | accuracy (%): 18.3 -> 28.5 | Improved |
| randomize camera height and distance (ablation) | Synthetic training · NTU CVS 0 deg | accuracy (%): 28.5 -> 31.3 | Improved |
Closest papers by meaning
- Generating Human Action Videos by Coupling 3D Game Engines and Probabilistic Graphical Models (IJCV 2019 · similarity 0.91)
- Procedural Generation of Videos to Train Deep Action Recognition Networks (CVPR 2017 · similarity 0.91)
- Learning Human Action Recognition Representations Without Real Humans (NeurIPS 2023 · similarity 0.90)
- Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition (IJCV 2019 · similarity 0.90)
- HumanVid: Demystifying Training Data for Camera-controllable Human Image Animation (NeurIPS 2024 · similarity 0.88)
- HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation (CVPR 2025 · 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.
