AI Paper Atlas

Research map / Image generation / Editing and style transfer

EDICT: Exact Diffusion Inversion via Coupled Transformations

Bram Wallace, Akash Gokul, Nikhil Naik

CVPR 2023 · Image generation › Editing and style transfer · approach: Inversion-based diffusion editing · 122 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 50 steps, unconditional reconstructionMSE: 0.030083 -> 0.015260Improved
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 50 steps, conditional reconstructionMSE: 0.418175 -> 0.015260Improved
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 200 steps, unconditional reconstructionMSE: 0.023406 -> 0.015260Improved
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 200 steps, conditional reconstructionMSE: 0.496944 -> 0.015260Improved
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 1000 steps, unconditional reconstructionMSE: 0.018960 -> 0.015260Improved
Use coupled transformations with two alternating noise vectors instead of DDIM inversionStable Diffusion 1.4 · MS-COCO-2017 validation, 1000 steps, conditional reconstructionMSE: 0.509345 -> 0.015260Improved
Add intermediate mixing layers rather than use coupling without mixing (ablation)EDICT · Text-conditioned image generationimage quality and compositionImproved
Use no intermediate mixing layers (ablation)EDICT · Text-conditioned image generationimage quality and compositionHurt
Set the mixing coefficient too high, such as p=0.97 (ablation)EDICT · 50-step generation and inversionlatent alignmentMixed
Set the mixing coefficient too low (ablation)EDICT · 50-step inversionlatent alignmentHurt
Increase the number of baseline editing steps from 50 to 100 or 250 (ablation)DDIM and other baseline editing methods · Image editingvisual edit qualityNo clear effect

Later papers that compared with this method

5 later accepted papers report EDICT as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy