Pith. sign in

REVIEW 5 cited by

VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2411.17451 v2 pith:FKH6QOCR submitted 2024-11-26 cs.CV cs.CL

VL-RewardBench: A Challenging Benchmark for Vision-Language Generative Reward Models

classification cs.CV cs.CL
keywords modelsvl-genrmsvl-rewardbenchtasksaccuracyvision-languagebenchmarkchallenging
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
abstract

Vision-language generative reward models (VL-GenRMs) play a crucial role in aligning and evaluating multimodal AI systems, yet their own evaluation remains under-explored. Current assessment methods primarily rely on AI-annotated preference labels from traditional VL tasks, which can introduce biases and often fail to effectively challenge state-of-the-art models. To address these limitations, we introduce VL-RewardBench, a comprehensive benchmark spanning general multimodal queries, visual hallucination detection, and complex reasoning tasks. Through our AI-assisted annotation pipeline that combines sample selection with human verification, we curate 1,250 high-quality examples specifically designed to probe VL-GenRMs limitations. Comprehensive evaluation across 16 leading large vision-language models demonstrates VL-RewardBench's effectiveness as a challenging testbed, where even GPT-4o achieves only 65.4% accuracy, and state-of-the-art open-source models such as Qwen2-VL-72B, struggle to surpass random-guessing. Importantly, performance on VL-RewardBench strongly correlates (Pearson's r $>$ 0.9) with MMMU-Pro accuracy using Best-of-N sampling with VL-GenRMs. Analysis experiments uncover three critical insights for improving VL-GenRMs: (i) models predominantly fail at basic visual perception tasks rather than reasoning tasks; (ii) inference-time scaling benefits vary dramatically by model capacity; and (iii) training VL-GenRMs to learn to judge substantially boosts judgment capability (+14.7% accuracy for a 7B VL-GenRM). We believe VL-RewardBench along with the experimental insights will become a valuable resource for advancing VL-GenRMs.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents

    cs.LG 2026-06 unverdicted novelty 7.0

    QVal is a new evaluation framework that directly measures dense supervision quality via Q-alignment to a reference policy, showing simple prompting baselines outperform 21 other methods across environments and models.

  2. Unified Reward Model for Multimodal Understanding and Generation

    cs.CV 2025-03 unverdicted novelty 7.0

    UnifiedReward is the first unified reward model that jointly assesses multimodal understanding and generation to provide better preference signals for aligning vision models via DPO.

  3. LLaVA-Critic-R1: Your Critic Model is Secretly a Strong Policy Model

    cs.CV 2025-08 conditional novelty 6.0

    RL training on preference-labeled critic data transforms a 7B vision-language model into both a stronger critic and a stronger generative policy, improving average benchmark accuracy by 5.7% and enabling self-critique...

  4. RewardBench 2: Advancing Reward Model Evaluation

    cs.CL 2025-06 unverdicted novelty 6.0

    RewardBench 2 is a new benchmark that supplies challenging fresh human prompts for reward model evaluation, yielding lower average scores but higher correlation with downstream best-of-N sampling and RLHF training per...

  5. Reinforcement Learning from Human Feedback

    cs.LG 2025-04 unverdicted novelty 2.0

    The book introduces the origins, mathematical setup, and optimization stages of RLHF including reward modeling, reinforcement learning, rejection sampling, and direct alignment algorithms.