detecting-ungrounded-object-mentions-open-ended-captions-white-box-vision-language-modelsFor detecting ungrounded object mentions in open-ended captions from white-box vision-language models, a single internal support signal (token likelihood, attention share, or image-text similarity) confuses support that comes from object-specific image evidence with support that comes from scene plausibility or the generated text prefix; scoring both whether image tokens carry object-specific evidence and whether support is more image-derived than prefix-derived separates these and raises detection AUROC.
Ingested from a paper but not yet reviewed by a human. It is deliberately inert: it does not move any technique’s standing, does not count toward the backtest, and is excluded anywhere a claim would carry weight. Read the source before relying on it.
Capability: Grounding answers in the image
Observed on
Object-existence hallucinations only, COCO/VOC category vocabularies, greedy-decoded detailed-description prompts, and access to internal hidden states; not tested on attributes, r.
Sources
- Object-level AUROC/AUPR on MSCOCO and Pascal VOC over eight LVLMs (LLaVA-1.5-7B/13B, LLaVA-NeXT, InstructBLIP, MiniGPT-4, InternVL3, Shikra, Qwen2.5-VL) against six training-free baselines; best average AUROC on both datasets. Ablations show compatibility alone at 72.25 AUROC rising to 83.88 with the evidence gate on LLaVA-1.5-7B, and counterfactuals (blank image, patch shuffle, prefix removal) move each component in the predicted direction. Single group, one prompt, one decoding setting; 500 images per benchmark.
Disagreeing is the most useful thing you can do here. Both sides of every contested claim in this catalog were assembled by the same person, which is its weakest point.
Related claims
- For detecting hallucinated text spans in vision-language model outputs, span localization and confidence calibration come from different architectures: a fine-tuned generative VLM proposes better span boundaries but scores them poorly, while a discriminative token tagger trained with a direct regression loss on human-consensus confidence gives better-correlated probabilities but conservative spans, so using the generative spans as proposals and the tagger's probabilities as scores beats either alone.Stating false facts confidently · unreviewed
- In LLaVA-1.5-7B captioning, restricting a LoRA adapter to attention heads picked by a hallucination diagnosis (attention-to-image drop around hallucinated object words, then ablation screening) reduces object hallucination, while an identically trained layer-matched random-head adapter does not — so the selection, not the added adapter capacity, carries the effect.Grounding answers in the image · unreviewed
- In diffusion vision-language models generating free-form captions, object tokens that are hallucinated commit at later denoising steps and with lower per-token confidence than grounded object tokens, so the commit-step trajectory is an available internal hallucination signal that has no counterpart in autoregressive decoding.Grounding answers in the image · unreviewed
- For open-weight agents where attention matrices are accessible, detecting injection by asking which context span actually drove the tool call — aggregating attention over the decision tokens and localizing the guiding span, then checking whether that span's provider has authority for the action — catches unauthorized tool invocations at higher true-positive and lower false-positive rates than static scanners that look for malicious-looking text, because benign-looking injected text is only harmful when it is what the model actually attends to.Following instructions hidden in data · unreviewed
- In 7B vision-language models that prune visual tokens during decoding for efficiency, hallucination rises as the generated caption gets longer, because pruning is irreversible and discards fine-grained evidence needed at later decoding steps; keeping the pruned tokens as a retrievable bank and pulling them back when next-token entropy spikes recovers the grounding without a second forward pass.Grounding answers in the image · unreviewed
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: A detector using a single internal support signal matching or beating the two-sided score across the same models and datasets, or ablations showing the evidence gate and the image-vs-prefix ratio add nothing over compatibility alone. Proposed technique, not catalogued: two-sided visual evidence and reliance scoring.