V-Zero: Answer-Label-Free On-Policy Distillation with Contrastive Evidence Gating for Fine-Grained Visual Reasoning

Hugging Face Daily Papers Papers
fine-grained-visual-reasoning multimodal distillation contrastive-learning on-policy-distillation visual-reasoning

Summary

V-Zero is a novel label-free framework for fine-grained visual reasoning that uses contrastive evidence gating and on-policy distillation to improve performance without annotated answer labels, achieving faster training than traditional methods.

Fine-grained visual reasoning requires multimodal large language models (MLLMs) to identify task-relevant visual evidence and ground their reasoning in local image regions. Existing agentic methods typically rely on reinforcement learning with verifiable rewards or supervised fine-tuning on large-scale annotated reasoning traces, leading to costly exploration, hand-designed verification rules, or heavy dependence on textual supervision. A natural way to avoid such external answer labels is to learn from trajectories sampled by the student itself, which points to On-Policy Distillation (OPD). To understand what OPD can and cannot provide for visual reasoning, we revisit it as negative-free stop-gradient alignment. This perspective shows that, although OPD provides effective token-level correction, its ceiling is constrained by the absence of trajectory-level discrimination. Motivated by these observations, we propose V-Zero, an answer-label-free framework for visual reasoning with contrastive evidence gating. V-Zero uses no annotated textual answer labels; instead, during training it pairs a question-relevant regional crop with a negative visual view to evaluate student-sampled trajectories and gate dense token-level distillation. Experiments on multiple visual reasoning benchmarks show that V-Zero consistently improves fine-grained visual reasoning while preserving strong generalization. Notably, V-Zero is more than 5times faster than previous supervised fine-tuning methods and more than 10times faster than reinforcement learning baselines. Code and dataset will be released at https://github.com/eVI-group-SCU/V-Zero
Original Article
View Cached Full Text

Cached at: 06/25/26, 05:17 AM

Paper page - V-Zero: Answer-Label-Free On-Policy Distillation with Contrastive Evidence Gating for Fine-Grained Visual Reasoning

Source: https://huggingface.co/papers/2606.25319

Abstract

A novel label-free framework for visual reasoning called V-Zero is presented, which uses contrastive evidence gating to improve fine-grained visual reasoning without requiring annotated answer labels, achieving faster training than traditional methods.

Fine-grained visual reasoningrequiresmultimodal large language models(MLLMs) to identify task-relevant visual evidence and ground their reasoning in local image regions. Existing agentic methods typically rely onreinforcement learningwith verifiable rewards orsupervised fine-tuningon large-scale annotated reasoning traces, leading to costly exploration, hand-designed verification rules, or heavy dependence on textual supervision. A natural way to avoid such external answer labels is to learn from trajectories sampled by the student itself, which points toOn-Policy Distillation(OPD). To understand what OPD can and cannot provide forvisual reasoning, we revisit it as negative-freestop-gradient alignment. This perspective shows that, although OPD provides effectivetoken-level correction, its ceiling is constrained by the absence oftrajectory-level discrimination. Motivated by these observations, we propose V-Zero, an answer-label-free framework forvisual reasoningwithcontrastive evidence gating. V-Zero uses no annotated textual answer labels; instead, during training it pairs a question-relevant regional crop with a negative visual view to evaluate student-sampled trajectories and gate dense token-level distillation. Experiments on multiplevisual reasoningbenchmarks show that V-Zero consistently improvesfine-grained visual reasoningwhile preserving strong generalization. Notably, V-Zero is more than 5times faster than previoussupervised fine-tuningmethods and more than 10times faster thanreinforcement learningbaselines. Code and dataset will be released at https://github.com/eVI-group-SCU/V-Zero

View arXiv pageView PDFProject pageGitHub14Add to collection

Get this paper in your agent:

hf papers read 2606\.25319

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

Models citing this paper1

#### hao05/v-zero-4b Visual Question Answering• 5B• Updated19 minutes ago • 21 • 1

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.25319 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.25319 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Visual Contrastive Self-Distillation

Hugging Face Daily Papers

VCSD removes the need for external teachers, privileged answers, or visual evidence in on-policy self-distillation by using content-erased control images to produce contrastive signals, consistently outperforming existing methods on vision-language benchmarks.

Evidence-RL: Towards Evidence-intensive Visual Reasoning

Hugging Face Daily Papers

This paper introduces Counterfactual Evidence Disentanglement (CED), a training-time method that makes vision-language models rely on concrete image evidence rather than language priors or shortcuts, improving visual reasoning grounding across benchmarks.

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.