diffusion-vision-language-models-generating-free-form-captions-object-tokens-hallucinated CLM-0044In 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.
Capability: Grounding answers in the image
Observed on
Two LaViDa backbones (LLaDA and Dream), 500 MSCOCO val2014 captions, 128 denoising steps, hallucination labelled by CHAIR object annotations..
Sources
- Measured: mean commit step 63.8 vs 37.6 (LaViDa-L) and 64.3 vs 42.2 (LaViDa-D) on a 128-step budget; commit-step ROC-AUC 0.699 and 0.667, PR-AUC 0.374 and 0.261 against base rates 0.190 and 0.152, 5-fold cross-validated. Only two backbones, one dataset, one caption prompt. An attention-based analysis found no difference, so the mechanism behind the signal is not established.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- In frozen large vision-language models, token-level hallucination evidence is distributed across the relational, spatial and sequential structure of the model's internal trace rather than contained in any single token's hidden state, so a detector that reads the attention graph over image patches, query and response tokens with interleaved GNN/ViT/GRU modules localizes hallucinated spans better than flat probes on the same activations.Grounding answers in the image · reviewed by AI
- 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 · reviewed by AI
- In 7B open vision-language models (LLaVA-1.5, Shikra), the per-token log-probability gap between a normal forward pass and one with visual features zeroed shrinks as a caption gets longer, and while that gap ranks hallucinated tokens slightly better than entropy or max-probability, its absolute separation is weak (AUROC around 0.57-0.59) and it fails entirely when aggregated to sentence level.Grounding answers in the image · reviewed by AI
- In 7-8B multimodal models, object hallucination during visually grounded token generation tracks a shift of the visual-language information ratio toward language inside mixed-modality "synergy" attention heads rather than a shortage of visual-specific heads, and rescaling visual versus language value vectors in those heads causally moves hallucination in both directions.Grounding answers in the image · reviewed by AI
- In 7B-class vision-language models (LLaVA-1.5, MiniGPT-4, mPLUG-Owl2), object hallucination is carried by an anisotropic set of directions in deep-layer hidden states, so attenuating each eigenmode in proportion to its hallucination-to-truthful variance ratio — estimated once offline from paired truthful/hallucinated captions and folded into the FFN output projection — lowers CHAIR rates, while a mean-offset subtraction or isotropic shrinkage of the same representations does not.Grounding answers in the image · reviewed by AI
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: On other diffusion VLMs or datasets, commit step for hallucinated object tokens is no later than for grounded ones, or ROC-AUC using commit step falls to chance. Proposed technique, not catalogued: Commit-step and confidence trajectory as a hallucination detector.