AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Is This Loss Informative? Faster Text-to-Image Customization by Tracking Objective Dynamics

Anton Voronov, Mikhail Khoroshikh, Artem Babenko, Max Ryabinin

NeurIPS 2023 · Image generation › Text-to-image diffusion · approach: Text-driven synthesis and semantic guidance · 0 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
early stopping with DVAR instead of full 6100 iterationsTextual Inversion · 48 conceptsadaptation time (min): 27.0 -> 3.1Improved
early stopping with DVAR instead of full 6100 iterationsTextual Inversion · 48 conceptsTrain CLIP img: 0.840 -> 0.795Hurt
early stopping with DVARTextual Inversion · 48 conceptsVal CLIP txt: 0.209 -> 0.227Improved
early stopping with DVARDreamBooth-LoRA · 48 conceptsadaptation time (min): 7.8 -> 4.9Improved
early stopping with DVARDreamBooth-LoRA · 48 conceptsVal CLIP img: 0.833 -> 0.687Hurt
early stopping with DVARDreamBooth-LoRA · 48 conceptsTrain CLIP img: 0.865 -> 0.784Hurt
early stopping with DVARCustom Diffusion · 48 conceptsiterations: 500 -> 348.1Improved
early stopping with DVARCustom Diffusion · 48 conceptsTrain CLIP img: 0.755 -> 0.742No clear effect
replace CLIP-s intermediate sampling with DVARTextual Inversion · 48 conceptsadaptation time (min): 9.6 -> 3.1Improved
replace CLIP-s intermediate sampling with DVARCustom Diffusion · 48 conceptsadaptation time (min): 9.7 -> 3.4Improved
evaluate loss on fixed deterministic batch rather than resampled training loss (ablation)Textual Inversion · convergence trackingloss informativenessImproved
resample diffusion timesteps in evaluation batch (ablation)L_det · convergence trackingloss informativenessHurt

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