# Rethinking CD: A Reproducibility Study and Extension on the Ineffectiveness of Contrastive Decoding at Mitigating Object Hallucinations in MLLMs

arXiv: https://arxiv.org/abs/2607.25196 · HTML: https://mlsanity.com/paper/2607.25196/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Hallucination detection and mitigation, one of 17 directions in LLM reasoning.
- Builds on 22 of its 24 references that are in the map.
- First posted on arXiv 28 Jul 2026.

## Placed in

- LLM reasoning / [Hallucination detection and mitigation](https://mlsanity.com/lr/hallucination-detection-and-mitigation/)
- Multimodal LLMs / vision-language models / [Hallucination mitigation in LVLMs](https://mlsanity.com/mllm/hallucination-mitigation-in-lvlms/)

## Builds on (most cited first, 22 of 22)

- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [SentencePiece: A simple and language independent subword tokenizer and detokenizer for Neural Text Processing](https://mlsanity.com/paper/1808.06226/) (2018)
- [Evaluating Object Hallucination in Large Vision-Language Models](https://mlsanity.com/paper/2305.10355/) (2023)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Understanding intermediate layers using linear classifier probes](https://mlsanity.com/paper/1610.01644/) (2016)
- [A-OKVQA: A Benchmark for Visual Question Answering using World Knowledge](https://mlsanity.com/paper/2206.01718/) (2022)
- [Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding](https://mlsanity.com/paper/2311.16922/) (2023)
- [Object Hallucination in Image Captioning](https://mlsanity.com/paper/1809.02156/) (2018)
- [Eliciting Latent Predictions from Transformers with the Tuned Lens](https://mlsanity.com/paper/2303.08112/) (2023)
- [Contrastive Decoding: Open-ended Text Generation as Optimization](https://mlsanity.com/paper/2210.15097/) (2022)
- [Efficiently Scaling Transformer Inference](https://mlsanity.com/paper/2211.05102/) (2022)
- [nocaps: novel object captioning at scale](https://mlsanity.com/paper/1812.08658/) (2018)
- [Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decoding](https://mlsanity.com/paper/2403.18715/) (2024)
- [HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding](https://mlsanity.com/paper/2403.00425/) (2024)
- [Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Models](https://mlsanity.com/paper/2408.02032/) (2024)
- [Plausible May Not Be Faithful: Probing Object Hallucination in Vision-Language Pre-training](https://mlsanity.com/paper/2210.07688/) (2022)
- [A Comprehensive Survey and Guide to Multimodal Large Language Models in Vision-Language Tasks](https://mlsanity.com/paper/2411.06284/) (2024)
- [Impact of Co-occurrence on Factual Knowledge of Large Language Models](https://mlsanity.com/paper/2310.08256/) (2023)
- [Investigating and Mitigating Object Hallucinations in Pretrained Vision-Language (CLIP) Models](https://mlsanity.com/paper/2410.03176/) (2024)
- [VaLiD: Mitigating the Hallucination of Large Vision Language Models by Visual Layer Fusion Contrastive Decoding](https://mlsanity.com/paper/2411.15839/) (2024)
- [The Mirage of Performance Gains: Why Contrastive Decoding Fails to Mitigate Object Hallucinations in MLLMs?](https://mlsanity.com/paper/2504.10020/) (2025)
- [D-Models and E-Models: Diversity-Stability Trade-offs in the Sampling Behavior of Large Language Models](https://mlsanity.com/paper/2601.17865/) (2026)

## Abstract (arXiv)

Multimodal large language models frequently describe objects that are not present in an image. Contrastive decoding has become a widely used inference-time remedy, and it is assumed to work by contrasting a normal forward pass against a hallucination-prone one so that hallucinated content is suppressed. However, whether the reported benchmark improvements actually arise from this mechanism has not been examined. Our key observation is that a correction which suppresses hallucination must depend on whether the model is hallucinating, and that this dependence can be measured directly inside the model. Following this idea, we analyse the internal computation of three contrastive decoding methods on both discriminative and generative tasks, and compare each against controls that remove the contrastive term while preserving its effect on the output. The correction turns out to be unrelated to hallucination at every layer, the gains in captioning come instead from a constraint that narrows sampling towards greedy decoding, and an unstructured perturbation of the same magnitude performs at least as well. Experiments across three models and standard hallucination benchmarks show that these improvements reflect a change in decoding behaviour rather than genuine hallucination mitigation.