Research map / 3D vision / 3D generation
View Synthesis by Appearance Flow
Tinghui Zhou, Shubham Tulsiani, Weilun Sun, Jitendra Malik, Alexei A. Efros
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 |
|---|---|---|---|
| predict appearance flow (copy input pixels) instead of generating pixels directly | single-view view synthesis CNN · ShapeNet car | mean pixel L1 error: 0.404 -> 0.368 | Improved |
| predict appearance flow instead of generating pixels directly | single-view view synthesis CNN · ShapeNet chair | mean pixel L1 error: 0.345 -> 0.323 | Improved |
| predict appearance flow instead of generating pixels directly | single-view view synthesis CNN · KITTI | mean pixel L1 error: 0.492 -> 0.471 | Improved |
| add a second input view with learned confidence-mask combination | appearance flow network · ShapeNet car | mean pixel L1 error: 0.368 -> 0.285 | Improved |
| add a second input view with confidence-mask combination | appearance flow network · ShapeNet chair | mean pixel L1 error: 0.323 -> 0.248 | Improved |
| add a second input view with confidence-mask combination | appearance flow network · KITTI | mean pixel L1 error: 0.471 -> 0.409 | Improved |
| add a second input view with hypothesis masks | Tatarchenko et al. direct pixel baseline · ShapeNet car | mean pixel L1 error: 0.404 -> 0.385 | Improved |
| add a second input view with hypothesis masks | Tatarchenko et al. direct pixel baseline · ShapeNet chair | mean pixel L1 error: 0.345 -> 0.334 | Improved |
| add a second input view with hypothesis masks | Tatarchenko et al. direct pixel baseline · KITTI | mean pixel L1 error: 0.492 -> 0.471 | No clear effect |
| predict appearance flow instead of direct pixels, judged by humans | single-view view synthesis CNN · ShapeNet car and chair user study | preference over baseline (%): 95% car, 93% chair | Improved |
| use appearance flow instead of pixel generation for views far from the input (e.g. front to side or rear) | single-view view synthesis CNN · ShapeNet cross-view confusion matrix | pixel L1 error | Hurt |
| add multi-scale reconstruction loss (ablation) | appearance flow network · preliminary experiments | gradient locality | Improved |
Closest papers by meaning
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.88)
- GenRec: Knowing Where to Reconstruct and Where to Generate (NeurIPS 2026 · similarity 0.86)
- LVSM: A Large View Synthesis Model with Minimal 3D Inductive Bias (ICLR 2025 · similarity 0.86)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.86)
- EscherNet: A Generative Model for Scalable View Synthesis (CVPR 2024 · similarity 0.86)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · 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.
