Say Less, Hallucinate Less? Diagnosing and Restoring Object Coverage in Preference-Optimized Vision-Language Models

Hansang LeearXiv 2026 · arXiv

In the quest to make Large Vision-Language Models (LVLMs) more reliable, researchers have turned to Preference Optimization (PO)—fine-tuning models on human or AI feedback to discourage "hallucinations" (mentioning objects not present in an image).

However, a new paper from Hansang Lee at Seoul Women’s University asks a provocative question: Are these models actually getting smarter, or are they just learning to shut up?

TL;DR

Preference Optimization (PO) reduces VLM hallucinations largely by making models stop writing sooner, which inadvertently kills object recall. By simply forcing the base model to stop early, researchers reproduced up to 86% of the hallucination reduction seen in state-of-the-art PO models, suggesting that current "alignment" might just be a fancy way of teaching models to say less.

What Problem Does This Paper Address?

Object hallucination is a major hurdle for LVLMs like LLaVA. While PO techniques (like DPO or RLAIF-V) successfully lower hallucination rates, they also tend to produce shorter, less descriptive captions.

The author investigates whether the "gains" reported by PO models are a genuine improvement in visual understanding or merely a side effect of reduced output length. If a model mentions fewer objects, it has fewer opportunities to be wrong. This creates a "brevity trap" where metrics like the hallucination rate improve, but the model's utility (its ability to describe everything in the image) declines.

Methods

The paper introduces a recall-controlled evaluation framework. Instead of comparing models at their default settings, the author compares them at the same level of object recall.

  1. EOS Logit Bias: To simulate "early stopping" in the base model, the author adds a positive bias to the end-of-sequence (EOS) token during decoding. This allows the base model to reach the same (lower) recall levels as the PO-tuned models.
  2. Negative EOS Bias Probe: Conversely, the author applies a negative bias to PO models to force them to write longer, testing if they can "recover" the objects they stopped mentioning.
  3. Base-Guided Object-Category Restoration: The author proposes a training-free intervention. It takes the output of a PO model and appends high-confidence object categories that were mentioned by the base model but missed by the PO model.
Figure 1. For RLAIF-V, CHAIR-DPO and CSR, early stopping of the base model reproduces much of the CHAIRi gain, while extending the PO captions recovers little of their lost recall. Hallucination against coverage for LLaVA-1.5-7B and four PO checkpoints on the development sets. The gray curve is the base model with a positive EOS logit bias, and the vertical gap between a star and the curve at the same recall is the stopping-controlled gain. The short colored paths show each PO model with a negative EOS bias of up to 10, dotted where more than 5 percent of the captions reach the 512-token limit. RLAIF-V, CHAIR-DPO and CSR barely move along their paths. The diamonds show cross-fitted restoration of each PO model aimed at the base recall, and the dashed arrows show the part of the PO gain that restoration gives up to recover the lost recall.

Datasets

The study evaluates four public PO checkpoints of LLaVA-1.5-7B (RLAIF-V, CHAIR-DPO, CSR, and HA-DPO) using:

  • CHAIR (Caption Hallucination Assessment with Image Relevance): Specifically using 500 random COCO images (CHAIR-500) and a separate held-out set of 500 images.
  • AMBER: A multi-dimensional hallucination benchmark.
  • POPE: A benchmark for yes/no object existence questions.
  • MMStar: To ensure general reasoning abilities aren't degraded.

Key Insights and Ideas

  • The "Say Less" Strategy: PO models name significantly fewer object categories (2.1–2.8 per caption) compared to the base model (3.2). Surprisingly, simple early stopping of the base model reproduces 58% to 86% of the hallucination reduction (CHAIRi gain) of three major PO checkpoints.
  • Lost Recall is Hard to Recover: When PO models are forced to write longer (via negative EOS bias), they don't start naming the missing objects. Instead, 94% to 96% of the additional text just repeats object categories they already named.
  • The Base Model Still "Sees" More: Even when a PO model misses an object, the base model’s caption for the same image often includes it. This suggests the information is present in the model's internal representation but is being filtered out by the PO tuning.
  • F1 Stagnation: While hallucination rates drop, the set-level object F1 score (which balances precision and recall) shows no significant gain for any PO model.
Figure 2. Stopping and base-guided restoration. (a) A positive EOS bias makes the base model stop earlier and name fewer objects, and this reproduces much of the CHAIRi gain of PO. Teal and red mark grounded and hallucinated objects for illustration only, and the EOS bias does not use the ground truth. (b) A negative EOS bias makes the PO caption longer, but the added text repeats object categories that the caption already names. See Fig. 3b for the detailed example. (c) Restoration keeps the base-only object categories whose base confidence is above τ and appends them to the PO caption. It does not check them against the image.

Key Results

  • Early Stopping Parity: On COCO, the base model with a positive EOS bias matched the hallucination performance of RLAIF-V and CHAIR-DPO while maintaining similar recall.
  • Restoration Success: The "Base-guided restoration" method brought three PO checkpoints back to within one point of the base model's recall. Crucially, it did this while maintaining a lower hallucination rate than the base model itself.
  • Quantitative Gains: For RLAIF-V, restoration lowered the distinct-object hallucination rate ($H_u$) by 5.24 points compared to the base model at the same recall level.

Limitations and Future Work

  • Scope: The study focuses on LLaVA-1.5-7B and greedy decoding. Results might differ for larger models or different architectures.
  • Attribute/Relation Hallucinations: The paper focuses on object presence. Hallucinations regarding colors, counts, or spatial relations were not the primary focus.
  • Fluency: The restoration method uses a simple template ("There is also a {object} in the image"), which may not be as fluent as a naturally generated caption.
  • Future Training: The author suggests that future PO training should specifically target "grounded category coverage" to ensure that reducing hallucinations doesn't come at the cost of descriptive richness.

Summary updated

Abstract

Preference optimization (PO) is the standard way to reduce object hallucination in large vision-language models. We find that four public PO checkpoints of LLaVA-1.5-7B also lower object coverage. They name fewer object categories, and their lower hallucination rates bring no significant gain in set-level object F1 on CHAIR. We therefore ask how much of the PO gain the base model can reproduce by stopping earlier. A positive end-of-sequence (EOS) logit bias lets the base model reach the recall of each PO model, and this early stopping alone reproduces 58 to 86 percent of the CHAIRi gain of three checkpoints on two disjoint COCO sets. Writing longer does not bring the lost categories back. Under greedy decoding, a negative EOS bias recovers only 2 to 9 percent of the lost CHAIR recall of these checkpoints, and 94 to 96 percent of the added object mentions repeat categories already named. On CHAIR, the base caption of the same image still names about half of the missed object categories. We therefore propose base-guided object-category restoration, which appends high-confidence base-only categories to the PO caption without training. With thresholds fixed on a development set, it brings three checkpoints within one point of the base recall on a held-out COCO set with a lower distinct-object hallucination rate than the base model. Against the recall-matched base stopping curve, its gain holds for RLAIF-V near the base recall and for RLAIF-V and CHAIR-DPO at an intermediate recall. We will release the code of the recall-controlled evaluation.
BibTeX
@misc{lee2026say,
  author = {Hansang Lee},
  title = {{Say Less, Hallucinate Less? Diagnosing and Restoring Object Coverage in Preference-Optimized Vision-Language Models}},
  year = {2026},
  eprint = {2611.23179},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.23179},
}
Figure