Research map / Learning theory and algorithms / Optimal transport
A Sliced Wasserstein Loss for Neural Texture Synthesis
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace the Gram loss with the Sliced Wasserstein loss | Neural texture synthesis by optimization · Texture synthesis | visual texture quality | Helped |
| Use the Gram loss instead of the Sliced Wasserstein loss | Neural texture synthesis by optimization · Texture synthesis | visual texture quality | Hurt |
| Use the Sliced Wasserstein loss instead of the Gram loss | Neural texture synthesis · Texture synthesis runtime | runtime overhead: ≈ 1.7-2.8 × | Hurt |
| Increase the number of random projection directions (ablation) | Sliced Wasserstein loss optimization · Texture optimization | convergence speed | Helped |
| Decrease the number of random projection directions (ablation) | Sliced Wasserstein loss optimization · Texture optimization | convergence speed | Hurt |
| Add fixed spatial tags as an extra feature dimension | Sliced Wasserstein loss · Texture synthesis with spatial constraints | reproduction of spatial structure | Helped |
| Use the contextual loss instead of the Gram loss | Texture synthesis by optimization · Texture synthesis | visual texture quality | Hurt |
| Use the contextual loss | Texture synthesis · Texture synthesis memory use | maximum feasible resolution: 100^2; they limit to 65^2 | Hurt |
| Use the REMD loss instead of the Gram loss | Texture synthesis by optimization · Texture synthesis | visual texture quality | Hurt |
| Remove REMD's two regularization losses (ablation) | REMD texture synthesis · Texture synthesis | visual texture quality | Hurt |
Closest papers by meaning
- Revisiting Sliced Wasserstein on Images: From Vectorization to Convolution (NeurIPS 2022 · similarity 0.89)
- Neural Texture Synthesis With Guided Correspondence (CVPR 2023 · similarity 0.88)
- Wasserstein of Wasserstein Loss for Learning Generative Models (ICML 2019 · similarity 0.87)
- Sliced Wasserstein Generative Models (CVPR 2019 · similarity 0.86)
- Augmented Sliced Wasserstein Distances (ICLR 2022 · similarity 0.86)
- Generalized rectifier wavelet covariance models for texture synthesis (ICLR 2022 · 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.