Research map / 3D vision / Face and shape
Unsupervised Disentanglement of Pose, Appearance and Background from Images and Videos
Ayşegül Dündar, Kevin J. Shih, Animesh Garg, Robert Pottorff, Andrew Tao, Bryan Catanzaro
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 foreground-background factorized rendering (ablation) | landmark model (temp) · Human3.6M | landmark error: 3.07 -> 2.73 | Improved |
| add foreground-background factorized rendering (ablation) | landmark model (temp,tps) · Human3.6M | landmark error: 2.86 -> 2.73 | Improved |
| add foreground-background factorized rendering (ablation) | landmark model (temp) · BBC Pose | keypoint accuracy: 73.3% -> 78.8% | Improved |
| add foreground-background factorized rendering (ablation) | landmark model (tps) · MAFL | landmark error: 4.34 -> 2.76 | Improved |
| add predicted blending masks instead of direct addition of fg and bg (ablation) | Ours (No Mask) · MAFL | landmark error: 2.88 -> 2.76 | Improved |
| add TPS perturbation to temporal perturbation in baseline | Baseline · BBC Pose | keypoint accuracy: 73.3% -> 73.4% | No clear effect |
| add TPS perturbation to temporal perturbation in baseline | Baseline · Human3.6M | landmark error: 3.07 -> 2.86 | Improved |
| add background factorization (fewer landmarks) (ablation) | landmark model · BBC Pose | accuracy vs number of landmarks | Improved |
| add foreground-background factorization to landmark model, 16 landmarks (ablation) | landmark model · Human3.6M | lowest landmark containment in foreground: 0.05 -> 8.2 | Improved |
| change coordinate origin from center to top left (ablation) | Ours · BBC Pose | keypoint accuracy: 78.78 -> 79.16 | No clear effect |
| add factorized rendering to video prediction (ablation) | landmark LSTM prediction · KTH | LPIPS, SSIM, PSNR | Improved |
| add factorized rendering to video prediction (ablation) | landmark LSTM prediction · BAIR | LPIPS, SSIM, PSNR | Improved |
Closest papers by meaning
- Unsupervised Learning of Object Landmarks through Conditional Image Generation (NeurIPS 2018 · similarity 0.90)
- Unsupervised Learning of Object Landmarks by Factorized Spatial Embeddings (ICCV 2017 · similarity 0.90)
- Object landmark discovery through unsupervised adaptation (NeurIPS 2019 · similarity 0.89)
- Unsupervised Learning of Object Landmarks via Self-Training Correspondence (NeurIPS 2020 · similarity 0.89)
- From Keypoints to Object Landmarks via Self-Training Correspondence: A novel approach to Unsupervised Landmark Discovery (TPAMI 2023 · similarity 0.88)
- Pose-Guided Self-Training with Two-Stage Clustering for Unsupervised Landmark Discovery (CVPR 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.
