Answer-Conditioned Chain-of-Thought Distillation for Few-Shot Industrial Vision with Small VLMs

• URL: http://arxiv.org/abs/2607.10666v1
• Date: Sun, 12 Jul 2026 09:23:05 GMT
• Title: Answer-Conditioned Chain-of-Thought Distillation for Few-Shot Industrial Vision with Small VLMs
• Abstract summary: We propose answer-conditioned chain-of-thought (CoT) distillation for rapidly adapting small vision-language models to new industrial tasks.<n>A frontier VLM receives each training image along with its correct label and generates a justified visual explanation.<n>We validate on four industrial classification tasks spanning three image modalities using only 18 to 30 labeled images per task.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Deploying AI-based visual inspection in manufacturing is hard because requirements change often, new defect types appear, and large labeled datasets are rarely available. We propose answer-conditioned chain-of-thought (CoT) distillation for rapidly adapting small vision-language models (VLMs) to new industrial tasks using minimal labeled data. A frontier VLM receives each training image along with its correct label and generates a justified visual explanation. A 3B-parameter model is then fine-tuned on these reasoning-augmented examples via LoRA. By conditioning on correct answers, we ensure all training reasoning is directed toward the correct conclusion, which is critical because frontier models score as low as 24.1% on our hardest task. We validate on four industrial classification tasks spanning three image modalities using only 18 to 30 labeled images per task. Across 4 seeds per task (32 training runs), our method outperforms direct fine-tuning on all 16 seed-task combinations, with mean improvements of +1.7 to +4.4 percentage points. A controlled equal-budget experiment confirms the improvement comes from reasoning quality, not additional training steps. An unconditioned baseline demonstrates that with out answer-conditioning, wrong reasoning degrades performance by 17.8 percentage points. On weld radiograph classification, the fine-tuned 3B model outperforms GPT-4.1 by 10.0pp using just 24 training images.

Related papers

• HALDETECT at ImageEval 2026 Shared Tasks: Answer-First Contrastive Grounding with QLoRA [0.0]
  We present HALDETECT, a system for the English hallucination-detection track (Task 1b) of ImageEval 2026.<n>We frame the item as one contrastive decision, emit the answer before its explanation, and structure reasoning around colour/texture, shape/form, and context.<n>Our best submitted adapter fine-tunes Qwen2.5-VL-7B-Instruct with 4-bit QLoRA while freezing the vision encoder and reaches Contrastive Instability (CI) 0.035 on the 1,000-item test set.
  arXiv  Detail & Related papers  (2026-09-10T08:35:29Z)
• CEPO: RLVR Self-Distillation using Contrastive Evidence Policy Optimization [50.59956036193097]
  A model produces a correct solution under reinforcement learning with verifiable rewards (RLVR)<n>Every token receives the same reward signal regardless of whether it was a decisive reasoning step or a grammatical filler.<n>We propose Contrastive Evidence Policy Optimization (CEPO)<n>CEPO asks a sharper question at every token: "does the correct answer favor this token?" but "does the correct answer favor it while the wrong answer disfavors it?"
  arXiv  Detail & Related papers  (2026-05-19T06:46:19Z)
• 20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone [17.84750728317393]
  We show that data curation can lift performance by +11.7pp on average across 20 public VLM benchmarks.<n>At 2B, our curated model surpasses InternVL3.5-2B by 9.9pp at 17x less training compute.<n>Data curation is a high-leverage tool for building better VLMs, reaching near-frontier accuracy at up to 150x less training compute.
  arXiv  Detail & Related papers  (2026-05-12T01:51:03Z)
• Understanding vision transformer robustness through the lens of out-of-distribution detection [59.72757235382676]
  Quantization reduces memory and inference costs at the risk of performance loss.<n>We investigate the behaviour of quantized small-variant popular vision transformers (DeiT, DeiT3, and ViT) on common out-of-distribution (OOD) datasets.
  arXiv  Detail & Related papers  (2026-02-01T22:00:59Z)
• DatBench: Discriminative, Faithful, and Efficient VLM Evaluations [17.234602646114997]
  Empirical evaluation serves as the primary compass guiding research progress in foundation models.<n>We propose three desiderata that evaluations should satisfy: faithfulness to the modality and application, discriminability between models of varying quality, and efficiency in compute.<n>We release DatBench-Full, a cleaned evaluation suite of 33 datasets spanning nine VLM capabilities, and DatBench, a discriminative subset that achieves 13x average speedup.
  arXiv  Detail & Related papers  (2026-01-05T18:07:51Z)
• HoneyBee: Data Recipes for Vision-Language Reasoners [90.83745691506329]
  We introduce several data curation approaches and study their impacts on vision-language models (VLMs)<n>We analyze the effects of context (image and question pair) sources, implement targeted data interventions, and explore scaling up images, questions, and chain-of-thought (CoT) solutions.<n>Motivated by these insights, we introduce HoneyBee, a large-scale, high-quality CoT reasoning dataset with 2.5M examples consisting 350K image-question pairs.
  arXiv  Detail & Related papers  (2025-10-14T07:23:44Z)
• STELAR-VISION: Self-Topology-Aware Efficient Learning for Aligned Reasoning in Vision [24.162895928364062]
  We introduce STELAR-Vision, a training framework for topology-aware reasoning.<n>At its core is TopoAug, a synthetic data pipeline that enriches training with diverse topological structures.<n>On MATH-V and VLM-S2H, STELAR-Vision improves accuracy by 9.7% over its base model and surpasses the larger Qwen2VL-72B-Instruct by 7.3%.
  arXiv  Detail & Related papers  (2025-08-12T07:27:50Z)
• Fine-Grained Image Recognition from Scratch with Teacher-Guided Data Augmentation [40.72028191529961]
  Fine-grained image recognition (FGIR) aims to distinguish visually similar sub-categories within a broader class, such as identifying bird species.<n>Most existing FGIR methods rely on backbones pretrained on large-scale datasets like ImageNet.<n>We introduce a novel training framework, TGDA, that integrates data-aware augmentation with weak supervision via a fine-grained-aware teacher model.
  arXiv  Detail & Related papers  (2025-07-16T11:37:33Z)
• LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models [59.0256377330646]
  Lens is a benchmark with 3.4K contemporary images and 60K+ human-authored questions covering eight tasks and 12 daily scenarios.<n>This dataset intrinsically supports to evaluate MLLMs to handle image-invariable prompts, from basic perception to compositional reasoning.<n>We evaluate 15+ frontier MLLMs such as Qwen2.5-VL-72B, InternVL3-78B, GPT-4o and two reasoning models QVQ-72B-preview and Kimi-VL.
  arXiv  Detail & Related papers  (2025-05-21T15:06:59Z)
• Improve Vision Language Model Chain-of-thought Reasoning [86.83335752119741]
  Chain-of-thought (CoT) reasoning in vision language models (VLMs) is crucial for improving interpretability and trustworthiness. We show that training VLM on short answers does not generalize well to reasoning tasks that require more detailed responses.
  arXiv  Detail & Related papers  (2024-10-21T17:00:06Z)
• CPT: Colorful Prompt Tuning for Pre-trained Vision-Language Models [101.5066760592534]
  We present Cross-modal Prompt Tuning (CPT), a novel paradigm for tuning Vision-Language Models (VL-PTMs) CPT reformulates visual grounding into a fill-in-the-blank problem with color-based co-referential markers in image and text, maximally mitigating the gap. Comprehensive experimental results show that prompt tuned VL-PTMs outperform their fine-tuned counterparts by a large margin.
  arXiv  Detail & Related papers  (2021-09-24T08:07:29Z)

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.