Research map / Vision-language and multimodal / Human and embodied modeling
MACK: Multimodal Aligned Conceptual Knowledge for Unpaired Image-text Matching
Yan Huang, Yuming Wang, Yunan Zeng, Liang Wang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| replace prototypical region representation with random region representation (ablation) | pretrained general knowledge · Flickr30k | Rs: 106.2 -> 31.7 | Hurt |
| remove adjectives when recomputing word representations (ablation) | pretrained general knowledge · Flickr30k | Rs: 106.2 -> 101.5 | Hurt |
| replace max-mean pooling with mean-mean pooling (ablation) | pretrained general knowledge · Flickr30k | Rs: 106.2 -> 62.2 | Hurt |
| add fine-tuning on unpaired images and texts with cycle-consistent loss | MACK · Flickr30k | Rs: 106.2 -> 153.5 | Improved |
| remove implicit one-to-one prior (ablation) | fine-tuned domain knowledge · Flickr30k | Rs: 153.5 -> 151.1 | Hurt |
| remove knowledge as cross-modal bridge (ablation) | fine-tuned domain knowledge · Flickr30k | Rs: 153.5 -> 3.2 | Hurt |
| re-rank CLIP with MACK | CLIP zero-shot · Flickr30k and MSCOCO | Rs: 525.6 -> 530.0; 386.1 -> 392.8 | Improved |
| re-rank ALBEF with MACK | ALBEF zero-shot · Flickr30k and MSCOCO | Rs: 506.9 -> 513.3; 428.3 -> 430.9 | Improved |
| re-rank VSRN with MACK | VSRN cross-dataset · MSCOCO -> Flickr30k, Flickr30k -> MSCOCO | Rs: 409.9 -> 411.7; 198.4 -> 201.4 | Improved |
| re-rank SAEM with MACK | SAEM cross-dataset · MSCOCO -> Flickr30k, Flickr30k -> MSCOCO | Rs: 415.5 -> 417.9; 219.8 -> 223.4 | Improved |
| raise max regions per prototype f from 20 to infinity (ablation) | ALBEF + MACK · Flickr30k | Rs: 509.8 -> 513.3 | Improved |
| raise max regions per prototype f from 20 to infinity (ablation) | ALBEF + MACK · MSCOCO | Rs: 427.5 -> 430.9 | Improved |
Closest papers by meaning
- MASS: Overcoming Language Bias in Image-Text Matching (AAAI 2025 · similarity 0.88)
- Identification of Necessary Semantic Undertakers in the Causal View for Image-Text Matching (AAAI 2024 · similarity 0.87)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.87)
- ARK: A Dual-Axis Multimodal Retrieval Benchmark along Reasoning and Knowledge (NeurIPS 2026 · similarity 0.87)
- CoMat: Aligning Text-to-Image Diffusion Model with Image-to-Text Concept Matching (NeurIPS 2024 · similarity 0.87)
- With Limited Data for Multimodal Alignment, Let the STRUCTURE Guide You (NeurIPS 2025 · similarity 0.87)
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.
