Research map / Image generation / Text-to-image diffusion
SpaText: Spatio-Textual Representation for Controllable Image Generation
Read the paperCheck 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 |
|---|---|---|---|
| Use SpaText's latent-based spatio-textual conditioning rather than Make-A-Scene | Text-to-image generation with sparse local prompts · COCO validation-derived inputs | FID: 6.7721 vs 21.367 | Helped |
| Use SpaText's latent-based spatio-textual conditioning rather than Make-A-Scene | Text-to-image generation with sparse local prompts · COCO validation-derived inputs | Local IOU: 0.2842 vs 0.2984 | Hurt |
| Use SpaText's latent-based model instead of No Token Left Behind | Text-to-image generation with sparse local prompts · COCO validation-derived inputs | Inference time (sec/image): 7 vs 326 | Helped |
| Use latent-based rather than pixel-based SpaText | SpaText · COCO validation-derived inputs | FID: 6.7721 vs 23.128 | Helped |
| Replace CLIP spatio-textual features with a binary mask (ablation) | SpaText · COCO validation-derived inputs | FID: 6.7721 vs 7.6085 | Hurt |
| Replace CLIP spatio-textual features with a binary mask (ablation) | SpaText · COCO validation-derived inputs | Local IOU: 0.2842 vs 0.2973 | Mixed |
| Remove the prior model and use CLIP text embeddings directly (ablation) | SpaText · COCO validation-derived inputs | FID: 6.7721 vs 7.025 | Hurt |
| Remove the prior model and use CLIP text embeddings directly (ablation) | SpaText · COCO validation-derived inputs | Local IOU: 0.2842 vs 0.3092 | Mixed |
| Use multi-scale instead of single-scale inference (ablation) | SpaText · COCO validation-derived inputs | FID: 6.7721 vs 10.5896 | Hurt |
| Use multi-scale instead of single-scale inference (ablation) | SpaText · COCO validation-derived inputs | Local distance: 0.7795 vs 0.7794 | Helped |
| Assign random labels to Make-A-Scene segments rather than ground-truth labels (ablation) | Make-A-Scene · COCO validation-derived inputs | FID: 21.367 vs 29.593 | Hurt |
Closest papers by meaning
- SSMG: Spatial-Semantic Map Guided Diffusion Model for Free-Form Layout-to-Image Generation (AAAI 2024 · similarity 0.89)
- Brush Your Text: Synthesize Any Scene Text on Images via Diffusion Model (AAAI 2024 · similarity 0.89)
- LLM Blueprint: Enabling Text-to-Image Generation with Complex and Detailed Prompts (ICLR 2024 · similarity 0.88)
- Layout-Agnostic Scene Text Image Synthesis with Diffusion Models (CVPR 2024 · similarity 0.88)
- GeoDiffusion: Text-Prompted Geometric Control for Object Detection Data Generation (ICLR 2024 · similarity 0.88)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · 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.