Research map / Image generation / Text-to-image diffusion
GASS: Geometry-Aware Spherical Sampling for Disentangled Diversity Enhancement in Text-to-Image Generation
Ye Zhu, Kaleb Newman, Johannes F. Lutzeyer, Adriana Romero-Soriano, Michal Drozdzal, Olga Russakovsky
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 re-normalization of perturbed embeddings (ablation) | GASS (SD3-M) · DrawBench | ImageReward: 0.732 -> 0.778 | Improved |
| add re-normalization of perturbed embeddings (ablation) | GASS (SD3-M) · DrawBench | ClipScore: 0.313 -> 0.320 | Improved |
| use dominant residual basis instead of isotropic perturbation (ablation) | GASS (SD3-M) · DrawBench | ClipScore: 0.308 -> 0.320 | Improved |
| use dominant residual basis instead of random orthogonal direction (ablation) | GASS (SD3-M) · DrawBench | ClipScore: 0.313 -> 0.320 | Improved |
| expand along both prompt-dependent and independent axes instead of only independent (ablation) | GASS (SD3-M) · DrawBench | SPP: 0.111 -> 0.114 | Improved |
| increase expansion range from 0.02 to 0.05 (ablation) | GASS (SD3-M) · DrawBench | SPP: 0.114 -> 0.112 | Hurt |
| use uniform instead of consecutive steps (t=10) (ablation) | GASS (SD3-M) · DrawBench | ImageReward: 0.808 -> 0.757 | Hurt |
| increase uniform guidance steps from 10 to 20 (ablation) | GASS (SD3-M) · DrawBench | Vendi Score: 8.127 -> 8.212 | Improved |
| add GASS to CFG on long prompts | SD3-M · DrawBench long prompts | Vendi Score: 7.549 -> 7.935 | Improved |
| add GASS to CFG on short prompts | SD3-M · DrawBench short prompts | ImageReward: 0.748 -> 0.698 | Hurt |
| add GASS to CFG | SD3-M · ImageNet-1K | Vendi Score: 28.119 -> 28.877 | Improved |
| add GASS to CFG | SD2.1 · ImageNet-1K | Density: 1.130 -> 1.015 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CADS — CADS: Unleashing the Diversity of Diffusion Models through Condition-Annealed Sampling (ICLR 2024) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TokenRepel: Generating Diverse Image Sets Without Sacrificing Quality (NeurIPS 2026 · similarity 0.88)
- Breaking the Lock-in: Diversifying Text-to-Image Generation via Representation Modulation (ICML 2026 · similarity 0.88)
- Diverse Text-to-3D Synthesis with Augmented Text Embedding (ECCV 2024 · similarity 0.87)
- HEART: Hyperspherical Embedding Alignment via Kent-Representation Traversal in Diffusion Models (NeurIPS 2026 · similarity 0.86)
- Let 2D Diffusion Model Know 3D-Consistency for Robust Text-to-3D Generation (ICLR 2024 · similarity 0.86)
- Minority-Focused Text-to-Image Generation via Prompt Optimization (CVPR 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.
