MLLMCLIP: Feature-Level Distillation of MLLM for Robust Vision-Language Representations

• URL: http://arxiv.org/abs/2608.25575v1
• Date: Wed, 26 Aug 2026 09:37:29 GMT
• Title: MLLMCLIP: Feature-Level Distillation of MLLM for Robust Vision-Language Representations
• Abstract summary: Pretrained vision-language models such as CLIP excel at zero-shot recognition but often fail at compositionality.<n>We propose MLLMCLIP, a framework that transfers multimodal knowledge directly from a generative Multimodal Large Language Model into a discriminative CLIP student.<n>MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval.
• Score: 35.542111485761744
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Pretrained vision-language models such as CLIP excel at zero-shot recognition but often fail at compositionality, particularly attribute-object and relational structures. Recent studies mitigate this issue by augmenting training with synthetic hard negatives generated by a cascade of large language models and text-to-image models, which incurs substantial pipeline overhead. We instead propose MLLMCLIP, a heterogeneous distillation framework that transfers multimodal knowledge directly from a generative Multimodal Large Language Model (MLLM) teacher into a discriminative CLIP student, bypassing synthetic data entirely. To bridge the architectural mismatch between the two paradigms, we introduce an attention-based per-layer token selection and a CKA-based distillation loss. Compared to prior CLIP-enhancement methods, MLLMCLIP achieves state-of-the-art compositional accuracy while delivering consistent gains on standard zero-shot classification and image-text retrieval, showing that feature-level distillation strengthens both compositional and general vision-language representation capability.

Related papers

• Growing Visual Generative Capacity for Pre-Trained MLLMs [60.826355079902505]
  Bridge is a pure autoregressive unified MLLM that augments pre-trained visual understanding models with generative ability.<n>We propose a semantic-to-pixel discrete representation that integrates compact semantic tokens with fine-grained pixel tokens.
  arXiv  Detail & Related papers  (2025-10-02T00:40:02Z)
• VITRIX-CLIPIN: Enhancing Fine-Grained Visual Understanding in CLIP via Instruction Editing Data and Long Captions [16.90061119174727]
  We present CLIP-IN, a novel framework that bolsters CLIP's fine-grained perception through two core innovations.<n> Firstly, we leverage instruction-editing datasets, originally designed for image manipulation, as a unique source of hard negative image-text pairs.<n> Secondly, CLIP-IN incorporates long captions, utilizing rotary positional encodings to capture rich semantic context often missed by standard CLIP.
  arXiv  Detail & Related papers  (2025-08-04T11:57:10Z)
• VladVA: Discriminative Fine-tuning of LVLMs [67.14293827774827]
  Contrastively-trained Vision-Language Models (VLMs) like CLIP have become the de facto approach for discriminative vision-language representation learning.<n>We propose to combine "the best of both worlds": a new training approach for discriminative fine-tuning of LVLMs.
  arXiv  Detail & Related papers  (2024-12-05T17:54:27Z)
• LLM2CLIP: Powerful Language Model Unlocks Richer Visual Representation [72.02635550088546]
  This work explores how large language models (LLMs) can enhance CLIP's capability, especially for processing longer and more complex image captions.<n>We introduce a caption-to-caption contrastive fine-tuning framework, significantly enhancing the discriminative quality of LLM outputs.<n>Our approach outperforms LoRA-based methods, achieving nearly fourfold faster training with superior performance.
  arXiv  Detail & Related papers  (2024-11-07T18:59:16Z)
• LLaVA-KD: A Framework of Distilling Multimodal Large Language Models [72.68665884790002]
  We propose a novel framework to transfer knowledge from l-MLLMs to s-MLLMs.<n>We introduce Multimodal Distillation (MDist) to transfer teacher model's robust representations across both visual and linguistic modalities.<n>We also propose a three-stage training scheme to fully exploit the potential of the proposed distillation strategy.
  arXiv  Detail & Related papers  (2024-10-21T17:41:28Z)
• Enhancing Fine-Grained Image Classifications via Cascaded Vision Language Models [0.0]
  This paper introduces CascadeVLM, an innovative framework that overcomes the constraints of previous CLIP-based methods. Experiments across various fine-grained image datasets demonstrate that CascadeVLM significantly outperforms existing models.
  arXiv  Detail & Related papers  (2024-05-18T14:12:04Z)
• RAR: Retrieving And Ranking Augmented MLLMs for Visual Recognition [78.97487780589574]
  Multimodal Large Language Models (MLLMs) excel at classifying fine-grained categories. This paper introduces a Retrieving And Ranking augmented method for MLLMs. Our proposed approach not only addresses the inherent limitations in fine-grained recognition but also preserves the model's comprehensive knowledge base.
  arXiv  Detail & Related papers  (2024-03-20T17:59:55Z)
• Distilling Knowledge from Text-to-Image Generative Models Improves Visio-Linguistic Reasoning in CLIP [57.53087077735303]
  We introduce SDS-CLIP, a lightweight and sample-efficient distillation method to enhance CLIP's compositional visio-linguistic reasoning. Our approach fine-tunes CLIP using a distillation objective borrowed from large text-to-image generative models like Stable-Diffusion. On the challenging Winoground benchmark, SDS-CLIP improves the visio-linguistic performance of various CLIP models by up to 7%, while on the ARO dataset, it boosts performance by up to 3%.
  arXiv  Detail & Related papers  (2023-07-18T13:10:11Z)
• Prompting Language-Informed Distribution for Compositional Zero-Shot Learning [73.49852821602057]
  Compositional zero-shot learning (CZSL) task aims to recognize unseen compositional visual concepts. We propose a model by prompting the language-informed distribution, aka., PLID, for the task. Experimental results on MIT-States, UT-Zappos, and C-GQA datasets show the superior performance of the PLID to the prior arts.
  arXiv  Detail & Related papers  (2023-05-23T18:00:22Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.