Research map / Vision-language and multimodal / Cross-modal learning
mPLUG-2: A Modularized Multi-modal Foundation Model Across Text, Image and Video
Haiyang Xu, Qinghao Ye, Ming Yan, Yaya Shi, Jiabo Ye, Yuanhong Xu, Chenliang Li, Bin Bi, Qi Qian, Wei Wang, Guohai Xu, Ji Zhang and 3 more
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add task and modality instructions (ablation) | mPLUG-2 · MSRVTT-Ret. | average Recall@1/5/10: 72.8 -> 73.5 (+0.7) | Improved |
| Add task and modality instructions (ablation) | mPLUG-2 · MSVD-QA | accuracy: 54.1 -> 55.3 (+1.2) | Improved |
| Add task and modality instructions (ablation) | mPLUG-2 · MSRVTT-Cap. | CIDEr: 71.8 -> 72.4 (+0.6) | Improved |
| Use local temporal modeling instead of temporal self-attention (ablation) | dual-vision encoder · MSRVTT-Ret. | average Recall@1/5/10: 70.3 -> 73.5 (+3.2) | Improved |
| Use local temporal modeling instead of temporal self-attention (ablation) | dual-vision encoder · MSVD-QA | accuracy: 55.1 -> 55.3 (+0.2) | Improved |
| Use local temporal modeling instead of temporal self-attention (ablation) | dual-vision encoder · MSRVTT-Cap. | CIDEr: 71.1 -> 72.4 (+1.3) | Improved |
| Replace temporal self-attention with temporal convolution (ablation) | dual-vision encoder · MSRVTT-Ret. | average Recall@1/5/10: 70.3 -> 71.4 (+1.1) | Improved |
| Replace temporal self-attention with temporal convolution (ablation) | dual-vision encoder · MSVD-QA | accuracy: 55.1 -> 55.0 (-0.1) | Hurt |
| Replace temporal self-attention with temporal convolution (ablation) | dual-vision encoder · MSRVTT-Cap. | CIDEr: 71.1 -> 71.7 (+0.6) | Improved |
| Add universal layers (ablation) | CLIP-ViT-L/14 · ImageNet, CIFAR10, CIFAR100, Cars, DTD, SUN, Food101 | average accuracy: 89.4 -> 90.6 (+1.2) | Mixed |
| Add universal layers after joint training (ablation) | BERT_base · GLUE tasks and VQA test-dev | score/accuracy: SST-2: 92.5 -> 93.5; RTE: 82.3 -> 85.2; MRPC: 86.6 -> 87.3; QQP: 90.6 -> 91.3; MNLI: 86.2 -> 87.6; QNLI: 92.1 -> 93.2; V | Improved |
| Use mPLUG-2 instead of HiTeA | video-text retrieval · DiDeMo | R@1: 56.5 -> 56.4 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports mPLUG2 as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- On Scaling Up a Multilingual Vision and Language Model (CVPR 2024 · better on 0 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MVBench: A Comprehensive Multi-modal Video Understanding Benchmark (CVPR 2024 · similarity 0.89)
- LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment (ICLR 2024 · similarity 0.89)
- mPLUG-Owl3: Towards Long Image-Sequence Understanding in Multi-Modal Large Language Models (ICLR 2025 · similarity 0.89)
- mPLUG-Owl2: Revolutionizing Multi-modal Large Language Model with Modality Collaboration (CVPR 2024 · similarity 0.89)
- MMInference: Accelerating Pre-filling for Long-Context Visual Language Models via Modality-Aware Permutation Sparse Attention (ICML 2025 · similarity 0.89)
- Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis (CVPR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
