Research map / 3D vision / Human pose estimation
Human Synthesis and Scene Compositing
Mihai Zanfir, Elisabeta Oneata, Alin-Ionut Popa, Andrei Zanfir, Cristian Sminchisescu
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 learned displacement field instead of no displacement field (ablation) | HUSC-S · DeepFashion | IS: 3.3876 -> 3.6343 | Improved |
| Use a learned displacement field instead of no displacement field (ablation) | HUSC-S · DeepFashion | SSIM: 0.8035 -> 0.8049 | Improved |
| Add a fixed body-only displacement field (ablation) | HUSC-S without displacement field · DeepFashion | IS: 3.3876 -> 3.4541 | Improved |
| Add a fixed body-only displacement field (ablation) | HUSC-S without displacement field · DeepFashion | SSIM: 0.8035 -> 0.8016 | Hurt |
| Replace fixed body-only displacement with learned displacement extended to clothing and hair (ablation) | HUSC-S · DeepFashion | IS: 3.4541 -> 3.6343 | Improved |
| Replace fixed body-only displacement with learned displacement extended to clothing and hair (ablation) | HUSC-S · DeepFashion | SSIM: 0.8016 -> 0.8049 | Improved |
| Add full-method components, including deformable skip connections and target dense pose in decoding (ablation) | HUSC-S with learned displacement field · DeepFashion | IS: 3.6343 -> 3.6950 | Improved |
| Add full-method components, including deformable skip connections and target dense pose in decoding (ablation) | HUSC-S with learned displacement field · DeepFashion | SSIM: 0.8049 -> 0.8135 | Improved |
| Refine perturbed composites with L1 and VGG losses | Appearance Compositing · COCO validation (1000 images) | LPIPS, full image: 0.0971 ± 0.0596 -> 0.0588 ± 0.0267 | Improved |
| Refine perturbed composites with L1 and VGG losses | Appearance Compositing · COCO validation (1000 images) | LPIPS, foreground: 0.2569 ± 0.0811 -> 0.1317 ± 0.0514 | Improved |
| Add discriminator GAN loss to appearance compositing (ablation) | Appearance Compositing (L1+VGG) · COCO validation (1000 images) | LPIPS, full image: 0.0588 ± 0.0267 -> 0.0542 ± 0.0243 | Improved |
| Add discriminator GAN loss to appearance compositing (ablation) | Appearance Compositing (L1+VGG) · COCO validation (1000 images) | LPIPS, foreground: 0.1317 ± 0.0514 -> 0.1165 ± 0.0480 | Improved |
Closest papers by meaning
- HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion (ICLR 2024 · similarity 0.87)
- HumanRef: Single Image to 3D Human Generation via Reference-Guided Diffusion (CVPR 2024 · similarity 0.87)
- GaussianIP: Identity-Preserving Realistic 3D Human Generation via Human-Centric Diffusion Prior (CVPR 2025 · similarity 0.87)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.86)
- HOComp: Interaction-Aware Human-Object Composition (NeurIPS 2025 · similarity 0.86)
- Multi-focal Conditioned Latent Diffusion for Person Image Synthesis (CVPR 2025 · similarity 0.86)
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.
