Research map / Image generation / Text-to-image diffusion
Norm-guided latent space exploration for text-to-image generation
Dvir Samuel, Rami Ben-Ari, Nir Darshan, Haggai Maron, Gal Chechik
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 norm-prior-guided interpolation instead of SLERP | Stable Diffusion seed interpolation · ImageNet1k | Acc: 30.41 -> 51.59 | Improved |
| Use norm-prior-guided interpolation instead of SLERP | Stable Diffusion seed interpolation · ImageNet1k | FID: 18.44 -> 6.78 | Improved |
| Use a norm-prior-induced centroid instead of sphere projection | Stable Diffusion seed centroid · ImageNet1k | Acc: 40.81 -> 67.24 | Improved |
| Use a norm-prior-induced centroid instead of sphere projection | Stable Diffusion seed centroid · ImageNet1k | FID: 14.28 -> 5.48 | Improved |
| Initialize SeedSelect with NAO-centroid seeds | SeedSelect · ImageNet1k in LAION2B, total | Total Acc: 91.3 -> 94.3 | Improved |
| Initialize SeedSelect with NAO-centroid seeds | SeedSelect · ImageNet1k in LAION2B, total | optimization time (sec): 298 -> 29 | Improved |
| Use NAO-centroid augmentation instead of SeedSelect | Few-shot classifier · miniImageNet, 5-way 5-shot | accuracy: 92.08±0.7 -> 93.21±0.6 | Improved |
| Use NAO-centroid augmentation instead of SeedSelect | Few-shot classifier · CIFAR-FS, 5-way 5-shot | accuracy: 94.87±0.4 -> 94.85±0.5 | No clear effect |
| Use NAO-centroid augmentation instead of SeedSelect | Few-shot classifier · CUB, 5-way 5-shot | accuracy: 96.01±0.4 -> 97.92±0.2 | Improved |
| Use NAO-centroid augmentation instead of SeedSelect | Long-tail classifier · iNaturalist | accuracy: 74.8±0.7 -> 75.3±0.7 | No clear effect |
| Use NAO-centroid augmentation instead of SeedSelect | Long-tail classifier · ImageNet-LT | accuracy: 74.9±0.5 -> 78.9±0.1 | Improved |
| Initialize SeedSelect with NAO seeds rather than random seeds | SeedSelect · few-shot recognition | accuracy | Improved |
Closest papers by meaning
- Generating Images of Rare Concepts Using Pre-trained Diffusion Models (AAAI 2024 · similarity 0.89)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · similarity 0.89)
- Progressive Text-to-Image Diffusion with Soft Latent Direction (AAAI 2024 · similarity 0.88)
- Würstchen: An Efficient Architecture for Large-Scale Text-to-Image Diffusion Models (ICLR 2024 · similarity 0.88)
- Kaleido Diffusion: Improving Conditional Diffusion Models with Autoregressive Latent Modeling (NeurIPS 2024 · similarity 0.88)
- Minority-Focused Text-to-Image Generation via Prompt Optimization (CVPR 2025 · similarity 0.88)
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.
