ADOPD: Reference-Privileged On-Policy Distillation for MLLM-Based Industrial Anomaly Detection

• URL: http://arxiv.org/abs/2608.09789v1
• Date: Mon, 10 Aug 2026 16:12:39 GMT
• Title: ADOPD: Reference-Privileged On-Policy Distillation for MLLM-Based Industrial Anomaly Detection
• Abstract summary: Industrial anomaly detection requires identifying fine-grained deviations from normal visual patterns.<n>Access to references during training allows a reference-aware teacher to supervise a query-only student.<n>We propose ADOPD, a reference-privileged on-policy distillation framework.
• Score: 15.5664235470661
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Industrial anomaly detection (IAD) requires identifying fine-grained deviations from normal visual patterns. Multimodal large language models (MLLMs) can improve recognition accuracy by comparing query images with references at inference time, but these benefits rely on additional retrieval and processing. We investigate whether the benefits of reference comparison can instead be internalized in the model parameters. Access to references during training allows a reference-aware teacher to supervise a query-only student. However, the teacher may favor plausible responses based on query cues or language priors rather than valid visual information. We propose ADOPD, a reference-privileged on-policy distillation framework. The teacher evaluates student-generated rollouts under matched and mismatched references. The matched-reference teacher-to-student log-ratio defines the token-level learning direction, specifying what the student should learn. The likelihood gap between the two reference views estimates reference-specific support and calibrates the sequence-level weight. ADOPD achieves 77.31% average accuracy on the MMAD benchmark under zero-shot inference, improving the Qwen3-VL-4B backbone by 6.14 points and outperforming its one-shot setting by 2.64 points. Experiments show that ADOPD learns a fine-grained anomaly inspection strategy from reference comparison. The project will be available at https://github.com/withTai/ADOPD.

Related papers

• V-Rubrics: Visual Faithfulness via Rubric-Based Reinforcement Learning [59.47823383593251]
  Vision-sampling models can produce fluent answers that are insufficiently grounded in the visual evidence.<n>We introduce Visuals-Based Reinforcement Learning, which decomposes reference responses into atomic propositions.<n>We train our model based on the same SFT checkpoint using component-wise, prefix-localized rubric credit.
  arXiv  Detail & Related papers  (2026-08-26T09:40:46Z)
• Rethinking Privileged Information in On-Policy Self-Distillation [45.88028371034407]
  On-policy self-distillation (OPSD) trains a student on its own responses using token-level supervision from the same model conditioned on privileged reference information.<n>We investigate whether performance gains from OPSD show that the student learned the information in the reference or instead reflect recovery of reasoning behavior already present in the base model.
  arXiv  Detail & Related papers  (2026-08-18T19:42:16Z)
• Lightning OPD 2.0: Mitigating Style Bias in Cross-Teacher On-Policy Distillation for Large Reasoning Models [23.269235615604245]
  We introduce Lightning OPD 2.0 with cross-fitted style residualization.<n>Across mathematical reasoning and code generation benchmarks, Lightning OPD 2.0 consistently outperforms Lightning OPD in cross-teacher settings.
  arXiv  Detail & Related papers  (2026-07-30T16:17:15Z)
• Visual Contrastive Self-Distillation [61.298159957622886]
  We propose Visual Contrastive Self-Distillation, which converts image-content removal into an on-policy self-distillation signal.<n>We use this contrast to sharpen the teacher's original-image distribution within its plausible support, and distill the resulting full-distribution target into the student.
  arXiv  Detail & Related papers  (2026-07-23T17:37:37Z)
• H$^2$SD: Hybrid Hindsight Self-Distillation [61.71131241473028]
  Reinforcement learning with verifiable rewards (RLVR) provides reliable outcome supervision for language model reasoning.<n>Existing self-distillation methods add a privileged teacher but typically assign it a fixed role.<n>We introduce Hybrid Hindsight Self-Distillation ($mathrmH2mathrmSD$), which jointly adapts teacher context and update strategy to trajectory correctness.
  arXiv  Detail & Related papers  (2026-07-21T10:47:27Z)
• dOPSD: On-Policy Self-Distillation for Diffusion Language Models [52.60302464420127]
  Diffusion large language models generate text by iteratively denoising a masked sequence.<n>Supervised fine-tuning is off-policy and suffers from exposure bias, while reinforcement learning gives only sparse, sequence-level rewards.<n>On-policy self-distillation (OPSD) offers a promising alternative, using one model as both student and teacher.
  arXiv  Detail & Related papers  (2026-07-05T17:47:29Z)
• A Formula-Driven Survey and Research Agenda for On-Policy Distillation [4.397842507533513]
  This survey studies OPD as a feedback-to-update problem rather than a single loss family.<n>We develop a formula-driven taxonomy from two routes -- direct distributional losses and policy-gradient-style log-ratio updates.
  arXiv  Detail & Related papers  (2026-06-22T03:09:21Z)
• AVSD: Adaptive-View Self-Distillation by Balancing Consensus and Teacher-Specific Privileged Signals [65.71225905458182]
  Self-distillation enables language models to learn on-policy from their own trajectories by using the same model as both student and teacher.<n>This setup provides dense token-level feedback without relying on a separate external model.<n>We introduce AVSD, a novel method of self-distillation with multiple privileged-information views.
  arXiv  Detail & Related papers  (2026-05-20T03:06:36Z)
• RefOnce: Distilling References into a Prototype Memory for Referring Camouflaged Object Detection [29.264612094775988]
  Ref-COD segments specified camouflaged objects in a scene by leveraging a small set of referring images.<n>Current systems adopt a dual-branch design that requires reference images at test time.<n>We introduce a Ref-COD framework that distills references into a class-prototype memory during training and synthesizes a reference vector at inference.
  arXiv  Detail & Related papers  (2025-11-26T02:42:52Z)
• FineRef: Fine-Grained Error Reflection and Correction for Long-Form Generation with Citations [30.28908306106096]
  FineRef teaches the model to self-identify and correct two key citation errors, mismatch and irrelevance, on a per-citation basis.<n>FineRef significantly improves both citation performance and answer accuracy.<n>Our 7B model outperforms GPT-4 by up to 18% in Citation F1 and 4% in EM Recall.
  arXiv  Detail & Related papers  (2025-11-18T09:35:12Z)
• Look As You Think: Unifying Reasoning and Visual Evidence Attribution for Verifiable Document RAG via Reinforcement Learning [55.232400251303794]
  Look As You Think (LAT) is a reinforcement learning framework that trains models to produce verifiable reasoning paths with consistent attribution.<n>LAT consistently improves the vanilla model in both single- and multi-image settings, yielding average gains of 8.23% in soft exact match (EM) and 47.0% in IoU@0.5.
  arXiv  Detail & Related papers  (2025-11-15T02:50:23Z)
• Compute as Teacher: Turning Inference Compute Into Reference-Free Supervision [26.922922043969958]
  We propose turning exploration into supervision through Compute as Teacher (CaT)<n>CaT synthesizes a single reference from a group of parallel rollouts and then optimizing toward it.<n>As a test-time procedure, CaT improves Gemma 3 4B, Qwen 3 4B, and Llama 3.1 8B.
  arXiv  Detail & Related papers  (2025-09-17T17:59:42Z)
• Unbiased Learning to Rank with Query-Level Click Propensity Estimation: Beyond Pointwise Observation and Relevance [74.43264459255121]
  In real-world scenarios, users often click only one or two results after examining multiple relevant options.<n>We propose a query-level click propensity model to capture the probability that users will click on different result lists.<n>Our method introduces a Dual Inverse Propensity Weighting mechanism to address both relevance saturation and position bias.
  arXiv  Detail & Related papers  (2025-02-17T03:55:51Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.