Research map / Image generation / Editing and style transfer
DiffusionCLIP: Text-Guided Diffusion Models for Robust Image Manipulation
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 GAN inversion with deterministic DDIM inversion and diffusion-model fine-tuning | Text-guided face image reconstruction · Face image reconstruction | MAE: 0.062 -> 0.020 | Helped |
| Replace GAN inversion with deterministic DDIM inversion and diffusion-model fine-tuning | Text-guided face image reconstruction · Face image reconstruction | LPIPS: 0.127 -> 0.073 | Helped |
| Replace GAN inversion with deterministic DDIM inversion and diffusion-model fine-tuning | Text-guided face image reconstruction · Face image reconstruction | SSIM: 0.877 -> 0.914 | Helped |
| Use diffusion-model fine-tuning rather than StyleCLIP | Text-guided attribute manipulation · CelebA-HQ and LSUN-Church | Directional CLIP similarity: 0.13 -> 0.17; 0.13 -> 0.20 | Helped |
| Use diffusion-model fine-tuning rather than StyleGAN-NADA | Text-guided attribute manipulation · CelebA-HQ and LSUN-Church | Segmentation-consistency: 89.4% -> 93.7%; 73.2% -> 78.1% | Helped |
| Use diffusion-model fine-tuning rather than StyleGAN-NADA | Text-guided face attribute manipulation · CelebA-HQ | Face identity similarity: 0.42 -> 0.70 | Helped |
| Increase the return step from t0=300 to t0=400 (ablation) | Diffusion image reconstruction · Face image reconstruction | MAE: 0.020 -> 0.021 | Hurt |
| Increase the return step from t0=300 to t0=500 (ablation) | Diffusion image reconstruction · Face image reconstruction | MAE: 0.020 -> 0.022 | Hurt |
| Increase the return step from t0=300 to t0=600 (ablation) | Diffusion image reconstruction · Face image reconstruction | MAE: 0.020 -> 0.024 | Hurt |
| Use fewer forward diffusion steps, S_for=6 (ablation) | DiffusionCLIP · Reconstruction and identity preservation | Identity preservation | No clear effect |
| Use S_gen=6 instead of S_gen=40 with S_for=40 (ablation) | DiffusionCLIP · Reconstruction | High-frequency detail preservation | Hurt |
Closest papers by meaning
- Multi-Region Text-Driven Manipulation of Diffusion Imagery (AAAI 2024 · similarity 0.88)
- DragonDiffusion: Enabling Drag-style Manipulation on Diffusion Models (ICLR 2024 · similarity 0.88)
- DragDiffusion: Harnessing Diffusion Models for Interactive Point-based Image Editing (CVPR 2024 · similarity 0.88)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.88)
- NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models (CVPR 2024 · similarity 0.87)
- E$^2$GAN: Efficient Training of Efficient GANs for Image-to-Image Translation (ICML 2024 · similarity 0.87)
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.