Research map / Image generation / Editing and style transfer
Editable Image Elements for Controllable Synthesis
Jiteng Mu, Michaël Gharbi, Richard Zhang, Eli Shechtman, Nuno Vasconcelos, Xiaolong Wang, Taesung Park
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 staged training (content encoder first, then diffusion decoder) instead of joint training (ablation) | editable image elements · LAION held-out reconstruction | PSNR: 19.74 -> 22.98 | Improved |
| add staged training (joint training preferred only 34.2% of the time in editing) (ablation) | editable image elements · MTurk editing preference | preference vs ours (%): 34.2% | Improved |
| freeze the content encoder when training the diffusion decoder (ablation) | editable image elements · LAION held-out reconstruction | PSNR: 21.35 -> 22.98 | Improved |
| freeze the content encoder (ablation) | editable image elements · LAION held-out reconstruction | FID: 10.48 -> 10.82 | Mixed |
| add random partition when dropping image elements during training (ablation) | editable image elements · MTurk editing preference | preference vs ours (%): 27.3% | Improved |
| add random partition (ablation) | editable image elements · LAION held-out reconstruction | FID: 9.75 -> 10.82 | Mixed |
| replace SD conditioning with image-element diffusion decoder vs plain L2-trained AE bottleneck | reconstruction · Reconstruction 5,000 samples | FID: 55.65 -> 10.82 | Improved |
| ours vs adversarially trained VAE on reconstruction | editable image elements · Reconstruction 5,000 samples | FID: 5.96 -> 10.82 | Hurt |
Closest papers by meaning
- High-Fidelity Diffusion-Based Image Editing (AAAI 2024 · similarity 0.90)
- Object-Aware Inversion and Reassembly for Image Editing (ICLR 2024 · similarity 0.89)
- FunEditor: Achieving Complex Image Edits via Function Aggregation with Diffusion Models (AAAI 2025 · similarity 0.89)
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.89)
- Image Editing As Programs with Diffusion Models (NeurIPS 2025 · similarity 0.89)
- EditInfinity: Image Editing with Binary-Quantized Generative Models (NeurIPS 2025 · similarity 0.89)
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.
