Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic Scenes

Hugging Face Daily Papers Papers
multimodal-llms self-distillation spatial-grounding synthetic-data post-training on-policy

Summary

Where-OPD proposes spatially guided on-policy self-distillation for MLLMs, where a teacher with privileged spatial text guidance is distilled into a student using procedurally generated synthetic scenes, yielding a 3.23-point average gain on real-world perception benchmarks via synthetic-to-real transfer.

On-policy self-distillation has recently emerged as an effective approach for improving language-model reasoning by supervising students with a frozen or EMA version of themselves that receives privileged information. Its application to multimodal large language models (MLLMs), however, remains largely unexplored. Recent approaches use privileged visual information, such as image crops corresponding to a question, to improve fine-grained perception, but their gains are confined to tasks that benefit from such visual zooming and require either human-annotated grounding data or external teacher models. We introduce a different form of on-policy self-distillation for MLLMs that provides the teacher with textual, spatially grounded guidance identifying the visual elements relevant to a query. We use procedurally generated scenes with automatically available object identities and spatial coordinates, enabling scalable and annotation-free post-training. The teacher uses this spatial guidance to locate and integrate evidence from multiple relevant image regions, while the student learns to reproduce the resulting behavior from the image and question alone. Our approach consistently improves performance on counting, document and chart understanding benchmarks across multiple models. Importantly, although post-training uses only synthetic scenes, the resulting improvements transfer to real-world perception benchmarks, yielding a 3.23-point gain in average performance across CVBench, V*, ZoomBench, BLINK, HR-Bench, and MME-RealWorld. These results show that spatially grounded privileged information can induce broader perceptual capabilities through on-policy self-distillation, enabling substantial synthetic-to-real transfer beyond the task and data distribution used for post-training. Project page: https://github.com/sirkosophia/Where-OPD
Original Article
View Cached Full Text

Cached at: 10/02/26, 04:30 PM

Paper page - Where-OPD: Spatially Guided On-Policy Self-Distillation of MLLMs with Synthetic Scenes

Source: https://huggingface.co/papers/2610.02117

Abstract

On-policyself-distillationhasrecentlyemergedasaneffectiveapproachforimprovinglanguage-modelreasoningbysupervisingstudentswithafrozenorEMAversionofthemselvesthatreceivesprivilegedinformation.Itsapplicationtomultimodallargelanguagemodels(MLLMs),however,remainslargelyunexplored.Recentapproachesuseprivilegedvisualinformation,suchasimagecropscorrespondingtoaquestion,toimprovefine-grainedperception,buttheirgainsareconfinedtotasksthatbenefitfromsuchvisualzoomingandrequireeitherhuman-annotatedgroundingdataorexternalteachermodels.Weintroduceadifferentformofon-policyself-distillationforMLLMsthatprovidestheteacherwithtextual,spatiallygroundedguidanceidentifyingthevisualelementsrelevanttoaquery.Weuseprocedurallygeneratedsceneswithautomaticallyavailableobjectidentitiesandspatialcoordinates,enablingscalableandannotation-freepost-training.Theteacherusesthisspatialguidancetolocateandintegrateevidencefrommultiplerelevantimageregions,whilethestudentlearnstoreproducetheresultingbehaviorfromtheimageandquestionalone.Ourapproachconsistentlyimprovesperformanceoncounting,documentandchartunderstandingbenchmarksacrossmultiplemodels.Importantly,althoughpost-trainingusesonlysyntheticscenes,theresultingimprovementstransfertoreal-worldperceptionbenchmarks,yieldinga3.23-pointgaininaverageperformanceacrossCVBench,V*,ZoomBench,BLINK,HR-Bench,andMME-RealWorld.Theseresultsshowthatspatiallygroundedprivilegedinformationcaninducebroaderperceptualcapabilitiesthroughon-policyself-distillation,enablingsubstantialsynthetic-to-realtransferbeyondthetaskanddatadistributionusedforpost-training.Projectpage:https://github.com/sirkosophia/Where-OPD

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2610\.02117

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.02117 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.02117 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.02117 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

OPD-V: Visual On-Policy Self-Distillation with Modality Balance

Hugging Face Daily Papers

Introduces OPD-V, a visual on-policy self-distillation paradigm for multimodal large language models that leverages positive and negative teachers to exploit modality balance as privileged information, improving reasoning performance across benchmarks while reducing training cost.

Latent-MOPD: Latent Multi-Teacher On-Policy Distillation

arXiv cs.LG

The paper introduces Latent-MOPD, the first representation-level multi-teacher on-policy distillation method for LLMs, which integrates multiple specialist teachers through both their output distributions and hidden states without extra teacher training, outperforming token-only, representation-only, and averaging baselines across nine math, code, and logic benchmarks.

Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning

Hugging Face Daily Papers

This paper analyzes on-policy self-distillation (OPSD) for LLM reasoning and shows that unverified student scaffolds create an imitation gap that worsens with model scale, motivating OASIS, a method that supervises mostly verified by-label on-policy trajectories using model-generated attempts as teacher context, improving Qwen3 1.7B/4B/8B by 3.2-3.8 points on AIME/HMMT benchmarks.

On-Policy Self-Distillation without Any Supervision

Hugging Face Daily Papers

Introduces U-OPSD, an unsupervised on-policy self-distillation method that uses internal consistency and majority-vote pseudo-solutions to improve LLMs without external supervision, matching or exceeding supervised methods on math benchmarks.

Latent On-Policy Self-Distillation

Hugging Face Daily Papers

This paper introduces Latent On-Policy Self-Distillation (LOPD), a method that makes the teacher's privileged context learnable end-to-end from experience, providing dense token-level supervision to enhance agent performance and efficiency in agentic tool use and code generation.