2026· ITM Web of Conferences· 0 citations· 3 references
TL;DR
Overall, the results show that the baseline is fairly stable under lightweight prompt and decoding changes, however, these low-cost adjustments do not solve the model's main weakness, which still appears in spatial and compositional reasoning.
Abstract
Vision-language models now perform well on many image-text tasks, but they are still less reliable when a question requires multi-step reasoning. This paper examines that gap through a small but controlled reproduction experiment using Qwen2-VL on CLEVR. The study begins with environment verification and a small VQA v2 smoke test, and then moves to fixed CLEVR validation slices scored with exact-match accuracy, answer normalization, and simple error categories. On the first 100-sample slice, the zero-shot baseline reaches 96.0% accuracy. To test whether simple inference-time changes help, the study compares three prompt styles on the same slice. All three produce the same 96.0% result. A decoding ablation on a larger 200-sample slice shows that greedy decoding remains at 95.0% across response-length limits of 16, 32, and 64, while a sampled setting falls slightly to 94.5%. A final 500-sample confirmation run with the best simple setting, namely a constrained short-answer prompt and greedy decoding with a 16-token limit, reaches 94.4%. Overall, the results show that the baseline is fairly stable under lightweight prompt and decoding changes. However, these low-cost adjustments do not solve the model's main weakness, which still appears in spatial and compositional reasoning.
Vision-Language Models (VLMs) still struggle on tasks requiring complex visual understanding. We argue that the core issue is not high-level reasoning, but instead failing to locate critical details in the image. Due to this shortcoming, VLMs generate often plausible but incorrect reasoning based on flawed perceptual grounding. To address this, we propose Locator-Critic (LOCI), a training-free framework that decouples visual search from evidence verification. LOCI employs a Locator agent to propose candidate visual evidence and a separate Critic agent to evaluate its relevance and sufficiency. These agents engage in an iterative refinement loop, progressively improving the evidence until it is adequate to answer the given question. This decoupled, self-correcting process yields substantial performance gains, achieving state-of-the-art results on multiple complex visual benchmarks. LOCI improves accuracy for both open-weight models like Qwen3-VL (+12.1 on V*, +5.8 on HR-Bench and +11.2 on VisualProbe-Hard) and proprietary models like Gemini 2.5 Pro (+8.9 on V*, +4.3 on HR-Bench, +4.8 on VisualProbe-Hard).
Walid Bousselham, Mathilde Caron, Arsha Nagrani et al.· 0 citations
Large multimodal models tend to hallucinate visual detail fluently, which limits their deployment for fine-grained interpretation. We present HALDETECT, our system for the English hallucination-detection track (Task 1b) of ImageEval 2026, in which a system must identify, from an image and three culturally plausible statements, the single visually grounded one. We frame the item as one contrastive decision, emit the answer before its explanation, and structure reasoning around colour/texture, shape/form, and context. 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; we placed third of eight teams. Development experiments show that answer order can matter more than model scale and that adaptation beats prompting alone. Retrospective paired analysis of the released gold labels confirms the QLoRA gain over the best prompt but not the small gap between the devtest-selected and best-test adapters, and reseeding all four training sizes shows that the apparent data-scaling curve does not survive a seed change. The 35 residual errors are culturally plausible function, material, and recognition distinctions; naive adapter voting does not help.
Modality reliance in VLMs is not fixed, but varies across tasks, evidence structures, models, and evaluation settings, which shows that modality reliance in VLMs is not fixed, but varies across tasks, evidence structures, and evaluation settings.
Fine-tuning a pretrained LLM into a vision-language model (VLM) can erode the backbone's text capability, with the damage concentrated on tasks that require following exact output rules, such as instruction following, chain-of-thought reasoning graded on a strictly parsed final answer, and similar evaluations with strict graders. We trace this gap to attention-sink corruption: VL fine-tuning perturbs the early sink position that anchors a large fraction of attention probability, and how well the base LLM preserves its sink tracks how much of the affected capability survives adaptation. Building on this view, we introduce Sink Strength, a single scalar computed on the base LLM in a few seconds on a single GPU that predicts post-VL degradation without any VL training. It consistently tracks relative degradation across the six VLM-LLM pairs and multiple format-sensitive tasks. Complementing this diagnostic, we find that post-pretraining QK-RMSNorm injection fails to reproduce the protection of native QK-RMSNorm, while several off-the-shelf weight-merging settings fail to recover the lost capability after VL training. These negative results underscore the value of screening backbones with Sink Strength before VL training and narrow the intervention space toward head-selective training-time protection.
Minsik Choi, Geewook Kim, Young Geun Kim· 0 citations
VGI-bench is introduced, containing 27 tasks and 810 instances, organized by a two-level taxonomy of task domains and skill tags for fine-grained evaluation of visual reasoning capabilities of video generation models, and it is hoped VGI-bench will help stimulate the development of next-generation video generation models.
Thinking-Once is proposed, a training-free, single-visual-pass evidence-routing method that reconstructs question-conditioned attention at this window, preserves core entity tokens and compact background context, and routes this evidence to later layers without extra visual encoding.
Z. Mao, Xian-Jie Liu, Tian-Yu Meng et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
