arXiv:2605. 01733v3 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) hallucinate objects that are not present, and a growing line of work tries to curb this by feeding the model its own generated caption as auxiliary evidence -- assuming that a caption, once available, is something to consume.
Object hallucination in multimodal large language models arises when language priors and corpus co-occurrence bias outweigh the visual evidence, with nothing tying an individual object mention to what the image shows. Most remedies intervene at decoding time without training, yet under a unified protocol their benefit is confined to short captions;supervised fine-tuning (SFT) on a detail- rich corpus lengthens captions, but over forty percent still name absent objects.
The paper introduces Visual Retrieval Heads (VRHs), a small fraction of attention heads in vision‑language models that are causally responsible for grounding text descriptions to image regions. By recasting head‑scoring methods and evaluating across eleven VLMs and five benchmarks, the authors show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect. VRHs generalize across various visual reference tasks, preserve output format while corrupting localization, and transfer causally across models sharing an LLM backbone.
arXiv:2509. 22415v3 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) have achieved strong vision-language performance, yet their token-level visual evidence remains difficult to inspect.
The paper demonstrates that vision‑language models (VLMs) possess a small set of attention heads, called Visual Retrieval Heads (VRHs), that are causally responsible for linking text prompts to specific image regions. By adapting head‑scoring techniques from language models, the authors identify VRHs as the heads whose attention from output prediction tokens, summed over the ground‑truth referent region, most reliably indicates causal grounding. Experiments across eleven VLMs and five referring‑expression benchmarks show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect, and that VRHs generalize across diverse visual tasks and transfer across models sharing an LLM backbone.
arXiv:2607. 27763v1 Announce Type: cross Abstract: We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions.
arXiv:2606. 28551v1 Announce Type: cross Abstract: Building performant Vision-Language Models (VLMs) requires carefully curating large-scale training datasets, yet the community lacks systematic benchmarks for evaluating such curation strategies.
ET‑Prune is a training‑free framework that dynamically allocates visual token budgets in multimodal large language models based on question‑conditioned evidence. It protects text‑like spatial regions, converts evidence uncertainty into a token floor, and progressively prunes concentrated evidence while retaining diffuse or text‑dense tokens. In six backbone‑benchmark comparisons, ET‑Prune matches or outperforms other pruned methods while keeping roughly half the visual tokens, achieving notable gains on OCRBench‑v2 and MMBench v1.1.
arXiv:2607. 28986v1 Announce Type: cross Abstract: Zero-shot image captioning (ZIC) describes images without paired image-caption supervision during captioner training, relying on text-only corpora and frozen pretrained image-text scorers.
We describe the DS@GT submissions to the ImageCLEFmedical Caption 2026 challenge, which continues a long-running benchmark on the ROCOv2 dataset with two tracks: Concept Detection (Task 1), assigning UMLS Concept Unique Identifiers (CUIs) to radiology images, and Caption Prediction (Task 2), generating natural-language captions. For Task 1, our primary submission was a three-way late-fusion ensemble of ConvNeXt-V2, BiomedCLIP ViT-B/16, and DenseNet-169 with a regularized ''Honest Threshold Tuning'' procedure designed to avoid validation overfitting on rare concepts; this submission ranked first on the official submission with a primary $F_1$ of $0.
The paper introduces HN-CLIP, a new objective for dense-caption retrieval that adapts similarity margins per negative example using the text encoder’s own geometry. By adding a detached caption‑similarity matrix to the negative logits, HN‑CLIP addresses the issue of near‑duplicate captions that cause premature loss saturation in InfoNCE training. Experiments on four benchmarks show that HN‑CLIP outperforms leading methods by 2.4–4.3 R@1, trains 2.4× faster than GOAL and 5.4× faster than StructXLIP, and achieves the best full‑data baseline with only 20% of the training data.
arXiv:2609.00591v1 Announce Type: new Abstract: An image may be worth a thousand words, but most captioning models describe it in only a few. Modern vision-language models produce fluent high-level c...