AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Harmonizing Visual Representations for Unified Multimodal Understanding and Generation

Size Wu, Wenwei Zhang, Lumin Xu, Sheng Jin, Zhonghua Wu, Qingyi Tao, Wentao Liu, Wei Li, Chen Change Loy

ICCV 2025 · Vision-language and multimodal › Human and embodied modeling · approach: Vision-language pretraining · 77 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 MAR encoder instead of VQGAN or VAE as shared visual encoder for understanding (ablation)Harmon-0.5B · POPEaccuracy: 63.8 -> 86.1Improved
use MAR encoder instead of SigLIP for understanding (ablation)Harmon-0.5B · MMMUaccuracy: 35.2 -> 33.3No clear effect
include stage I vision-language alignment (ablation)Harmon-0.5B · MME-Pscore: 1003 -> 1148Improved
include stage III high-quality fine-tuning (ablation)Harmon-0.5B · MJHQ-30KFID: 16.85 -> 6.08Improved
include stage II comprehensive multimodal training (ablation)Harmon-0.5B · GenEvaloverall: 0.65 -> 0.71Improved
increase stage III image resolution from 256 to 512 (ablation)Harmon-0.5B · MJHQ-30KFID: 9.91 -> 6.08Improved
increase stage III image resolution from 256 to 512 (ablation)Harmon-0.5B · GenEvaloverall: 0.68 -> 0.71Improved
add understanding loss when training generation (unified vs single-task) (ablation)Harmon · GenEvaloverallImproved
co-train understanding and generation instead of understanding only (ablation)Harmon · Understanding benchmarksaccuracyNo clear effect
use a separate SigLIP encoder for understanding (Janus-like decoupled design) instead of a shared MAR encoder (ablation)Harmon · GenEvaloverallHurt
use Harmon-1.5B with shared MAR encoder versus Janus-Pro-1.5B on understandingHarmon-1.5B · MMBenchaccuracy: 75.5 -> 65.5Hurt
use Harmon shared MAR encoder for text-to-image generationHarmon-1.5B · GenEvaloverall: 0.73 -> 0.76Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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