The paper introduces a new training framework for Visual Question Answering that leverages counterfactual contrastive learning to mitigate language bias and improve out‑of‑distribution generalization. It comprises a three‑stage curriculum for stable optimization, an enhanced Batch‑Contrastive loss for discriminative feature learning, and two regularizers—Answer‑Contrastive and Gradient‑Discrepancy—to refine predictions and enforce causal visual grounding. The resulting model attains 61.64% accuracy on the bias‑sensitive VQA‑CP v2 benchmark while maintaining 62.80% on the standard VQA v2 dataset, achieving a small generalization gap of 1.16%.
arXiv:2606. 29984v1 Announce Type: new Abstract: Reinforcement Learning (RL) is an important paradigm for improving the reasoning capabilities of Vision-Language Models (VLMs).
arXiv:2609.13228v1 Announce Type: new Abstract: Vision Language Models (VLMs) should rely on visual evidence that directly determines the correct answer, but supervision for grounding visual reasonin...
arXiv:2608. 08021v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) should answer from concrete image evidence rather than language priors, dataset shortcuts, or irrelevant visual context.
arXiv:2610.33841v1 Announce Type: cross Abstract: When an image and its caption disagree, can a user reliably control which one a vision-language model (VLM) follows? Accuracy alone cannot answer thi...
arXiv:2610.16977v1 Announce Type: cross Abstract: Visual in-context learning (ICL) enables vision-language models (VLMs) to adapt at test time via demonstration examples, but its effectiveness depend...
The paper introduces GRIT, a training framework that adds visual grounding to reinforcement learning with verifiable rewards (RLVR) for vision-language models, addressing the issue of Correct-but-Visually-Misgrounded Reasoning (CVMR). CVMR occurs when models produce correct final answers while misperceiving visual information during intermediate reasoning steps, a problem that standard RLVR rewards cannot detect. By analyzing token-level attention patterns, the authors show that CVMR tokens differ from well-grounded tokens and use this insight to guide GRIT, which improves both reasoning accuracy and grounding fidelity. Additionally, the authors present GRIT-Bench, a new benchmark for evaluating text-rich reasoning across varying text lengths and reasoning depths.
arXiv:2608. 06938v1 Announce Type: cross Abstract: The visual reasoning ability of multimodal large language models (MLLMs) is crucial for downstream applications, particularly counter-commonsense reasoning, which requires models to reason beyond common assumptions.
arXiv:2606. 06890v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence.
The paper introduces S‑MME, a benchmark for diagnosing shortcut‑biased behavior in Multimodal Large Language Models (MLLMs) on long‑video question answering. It proposes CLEAR, a training‑free framework that estimates and mitigates bias by treating the full video‑question pair as a complete view and intervening on hidden states during inference. Experiments on five benchmarks demonstrate that CLEAR improves prediction accuracy and robustness across different MLLMs.
EDCT-Bench is a benchmark that evaluates the faithfulness of Vision‑Language Models (VLMs) by using Explanation‑Driven Counterfactual Testing (EDCT). EDCT extracts visual concepts from a model’s natural language explanation, applies minimal verified edits to those concepts, and checks whether the model’s answer and explanation remain consistent with the edited image. The benchmark covers three domains—knowledge‑intensive VQA, safety‑critical driving, and 3D spatial reasoning—and reveals significant faithfulness gaps in current VLMs, while also showing that EDCT‑generated counterfactuals can improve training.
arXiv:2606. 02168v1 Announce Type: cross Abstract: Compositional visual question answering (VQA) represents a challenging yet fundamental task that requires models to comprehend novel combinations of previously learned concepts.
The paper introduces Self-RefineVQA, a training‑free, inference‑time framework that lets a single Vision‑Language Model iteratively generate, critique, and refine its own answers for table visual question answering. By forming a closed self‑feedback loop guided by prompts, the method corrects common reasoning errors such as spatial misalignment and logical hallucinations without requiring additional labeled data or model updates. Experiments on tabular VQA benchmarks show consistent accuracy gains across monolingual and cross‑lingual settings, especially for lightweight models, while keeping inference cost linear and eliminating the need for expensive supervised fine‑tuning.