CPCF: A Cross-Prompt Contrastive Framework for Referring Multimodal Large Language Models

Lanyun Zhu, Deyi Ji, Tianrun Chen, Haiyang Wu, De Wen Soh, Jun Liu
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:79836-79849, 2025.

Abstract

Referring MLLMs extend conventional multimodal large language models by allowing them to receive referring visual prompts and generate responses tailored to the indicated regions. However, these models often suffer from suboptimal performance due to incorrect responses tailored to misleading areas adjacent to or similar to the target region. This work introduces CPCF, a novel framework to address this issue and achieve superior results. CPCF contrasts outputs generated from the indicated visual prompt with those from contrastive prompts sampled from misleading regions, effectively suppressing the influence of erroneous information outside the target region on response generation. To further enhance the effectiveness and efficiency of our framework, several novel designs are proposed, including a prompt extraction network to automatically identify suitable contrastive prompts, a self-training method that leverages unlabeled data to improve training quality, and a distillation approach to reduce the additional computational overhead associated with contrastive decoding. Incorporating these novel designs, CPCF achieves state-of-the-art performance, as demonstrated by extensive experiments across multiple benchmarks. Project page: https://lanyunzhu.site/CPCF/

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zhu25h, title = {{CPCF}: A Cross-Prompt Contrastive Framework for Referring Multimodal Large Language Models}, author = {Zhu, Lanyun and Ji, Deyi and Chen, Tianrun and Wu, Haiyang and Soh, De Wen and Liu, Jun}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {79836--79849}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhu25h/zhu25h.pdf}, url = {https://proceedings.mlr.press/v267/zhu25h.html}, abstract = {Referring MLLMs extend conventional multimodal large language models by allowing them to receive referring visual prompts and generate responses tailored to the indicated regions. However, these models often suffer from suboptimal performance due to incorrect responses tailored to misleading areas adjacent to or similar to the target region. This work introduces CPCF, a novel framework to address this issue and achieve superior results. CPCF contrasts outputs generated from the indicated visual prompt with those from contrastive prompts sampled from misleading regions, effectively suppressing the influence of erroneous information outside the target region on response generation. To further enhance the effectiveness and efficiency of our framework, several novel designs are proposed, including a prompt extraction network to automatically identify suitable contrastive prompts, a self-training method that leverages unlabeled data to improve training quality, and a distillation approach to reduce the additional computational overhead associated with contrastive decoding. Incorporating these novel designs, CPCF achieves state-of-the-art performance, as demonstrated by extensive experiments across multiple benchmarks. Project page: https://lanyunzhu.site/CPCF/} }
Endnote
%0 Conference Paper %T CPCF: A Cross-Prompt Contrastive Framework for Referring Multimodal Large Language Models %A Lanyun Zhu %A Deyi Ji %A Tianrun Chen %A Haiyang Wu %A De Wen Soh %A Jun Liu %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhu25h %I PMLR %P 79836--79849 %U https://proceedings.mlr.press/v267/zhu25h.html %V 267 %X Referring MLLMs extend conventional multimodal large language models by allowing them to receive referring visual prompts and generate responses tailored to the indicated regions. However, these models often suffer from suboptimal performance due to incorrect responses tailored to misleading areas adjacent to or similar to the target region. This work introduces CPCF, a novel framework to address this issue and achieve superior results. CPCF contrasts outputs generated from the indicated visual prompt with those from contrastive prompts sampled from misleading regions, effectively suppressing the influence of erroneous information outside the target region on response generation. To further enhance the effectiveness and efficiency of our framework, several novel designs are proposed, including a prompt extraction network to automatically identify suitable contrastive prompts, a self-training method that leverages unlabeled data to improve training quality, and a distillation approach to reduce the additional computational overhead associated with contrastive decoding. Incorporating these novel designs, CPCF achieves state-of-the-art performance, as demonstrated by extensive experiments across multiple benchmarks. Project page: https://lanyunzhu.site/CPCF/
APA
Zhu, L., Ji, D., Chen, T., Wu, H., Soh, D.W. & Liu, J.. (2025). CPCF: A Cross-Prompt Contrastive Framework for Referring Multimodal Large Language Models. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:79836-79849 Available from https://proceedings.mlr.press/v267/zhu25h.html.

Related Material