7b-class-vision-language-models-llava-1-5-minigpt-4-mplug-owl2-object-hallucination CLM-0536mechanismsingle paperReviewed 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.
Capability: Grounding answers in the image
Observed on
Deeper layers only (editing layers 4-12 barely helps); requires a calibration set of image-paired truthful and hallucinated captions; demonstrated on open 7B VLMs on CHAIR/POPE/MME.
Sources
- Validated by: CHAIR_S/CHAIR_I on 500 MSCOCO images over three seeds against greedy, beam search, DoLa, VCD, OPERA, MARINE, HALC and Nullu (e.g. MiniGPT-4 CHAIR_S 21.20 greedy, 17.67 HALC, 16.87 ours), plus POPE accuracy/F1 and MME Dolan-More profiles; ablations on MiniGPT-4 compare mean-distortion subtraction (CHAIR_S 22.0) and uniform shrinkage (24.0) against the full filter (13.0) at matched layers.. Measured on three VLMs across CHAIR, POPE and MME with seed variance reported, plus controlled ablations isolating anisotropy and layer depth. [truncated]
Either way, bring evidence: a link, or what you saw.
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 open-source vision-language models (LLaVA-1.5, Qwen-VL-Chat, Qwen2.5-VL, LLaVA-NeXT), a training-free decoding intervention that detects large layer-to-layer and step-to-step shifts in hidden states and nudges the diverging states back toward their prior value lowers object-hallucination rates, but the lower rate comes with fewer objects mentioned and shorter outputs.Grounding answers in the image · reviewed by AI
- 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 open-source vision-language models (LLaVA-1.5, Qwen-VL-Chat, Qwen2.5-VL, LLaVA-NeXT), a training-free decoding intervention that detects large layer-to-layer and step-to-step shifts in hidden states and nudges the diverging states back toward their prior value lowers object-hallucination rates, but the lower rate comes with fewer objects mentioned and shorter outputs.Grounding answers in the image · reviewed by AI
- In 4-7B open vision-language models (LLaVA-1.5-7B, Qwen2.5-VL-7B, InternVL-2.5-4B), how the cross-modal attention distribution over visual tokens shifts between adjacent decoder layers separates hallucinated from grounded object predictions better than the model's sensitivity to masking the attended region, and better than attention drift measured against a fixed first or last layer anchor.Grounding answers in the image · reviewed by AI
- In 7B-class LVLMs (LLaVA-1.5, MiniGPT-4, mPLUG-Owl2), hallucination directions in activation space are not a single mode: clustering hallucinated-minus-faithful hidden-state differences into several low-rank subspaces and weighting them per test image lowers object-hallucination rates slightly below a single global null-space projection, with the margin small (CHAIR_I 5.30 to 4.92 on LLaVA-1.5) and inconsistent on yes/no existence probing.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: An isotropic or mean-offset correction at the same deep layers matching the Wiener filter's CHAIR reduction, or the Wiener filter's CHAIR gain disappearing once caption length and object coverage are held constant.