AI Paper Atlas

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

Cycle Consistency as Reward: Learning Image-Text Alignment without Human Preferences

Hyojin Bahng, Caroline Chan, Fredo Durand, Phillip Isola

ICCV 2025 · Image generation › Text-to-image diffusion · approach: T2I evaluation and controllable adapters · 19 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 DreamSim instead of LPIPS for image similarity (ablation)CycleReward-I2T · DetailCaps-4870pairwise accuracy: 53.16 -> 58.02Improved
use DreamSim instead of CLIP for image similarity (ablation)CycleReward-I2T · DetailCaps-4870pairwise accuracy: 57.90 -> 58.02Improved
use SBERT instead of BERT for text similarity (ablation)CycleReward-T2I · DetailCaps-4870pairwise accuracy: 47.27 -> 51.74Improved
use SBERT instead of BERT for text similarity (ablation)CycleReward-T2I · GenAI-Benchpairwise accuracy: 55.20 -> 55.52Hurt
use Stable Diffusion 3 instead of SDXL-Turbo as text-to-image decoder (ablation)CycleReward-I2T · DetailCaps-4870pairwise accuracy: 56.42 -> 58.02Improved
use InternVL2-26B instead of LLaVA-1.5-13B as image-to-text decoder (ablation)CycleReward-T2I · DetailCaps-4870pairwise accuracy: 51.74 -> 57.21Improved
learn a reward model from cycle consistency instead of using raw scoresCycleReward-I2T · DetailCaps-4870pairwise accuracy: 56.46 -> 58.02Improved
jointly train on I2T and T2I preferencesCycleReward · DetailCaps-4870pairwise accuracy: 58.02 -> 60.50Improved
label ImageRewardDB with cycle consistency instead of human labelsImageReward backbone · GenAI-Benchpairwise accuracy: 56.70 -> 54.58No clear effect
DPO Qwen-VL-Chat with CyclePrefDB-I2TQwen-VL-Chat · DeCapBenchDCScore: 26.47 -> 30.63Improved
DPO Qwen-VL-Chat with CyclePrefDB-I2TQwen-VL-Chat · MME perceptionscore: 1460.2 -> 1485.7Improved
Diffusion DPO SD1.5 with CyclePrefDB-T2IStable Diffusion 1.5 · T2I-CompBench Spatialscore: 11.49 -> 16.55Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

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