Research map / Vision-language and multimodal / Cross-modal learning
Deep Cross-Modal Projection Learning for Image-Text Matching
Ying Zhang, Huchuan Lu
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 |
|---|---|---|---|
| add softmax classification loss to CMPM (ablation) | CMPM · CUHK-PEDES image-to-text | R@1: 44.02 -> 45.38 | Improved |
| add weight normalization to softmax classification (ablation) | CMPM+softmax · CUHK-PEDES text-to-image | R@1: 45.38 -> 47.12 | Improved |
| add weight normalization to softmax classification (ablation) | CMPM+softmax · CUHK-PEDES image-to-text | R@1: 55.14 -> 56.51 | Improved |
| add cross-modal projection on top of weight-normalized softmax (full CMPC) (ablation) | CMPM+softmax+normW · CUHK-PEDES text-to-image | R@1: 47.12 -> 49.37 | Improved |
| add CMPM matching loss to CMPC-only classification (ablation) | CMPC · CUHK-PEDES text-to-image | R@1: 42.25 -> 49.37 | Improved |
| add weight normalization to CMPC alone (ablation) | CMPC without CMPM · CUHK-PEDES text-to-image | R@1: 16.93 -> 42.25 | Improved |
| add CMPC loss to CMPM (ablation) | CMPM · CUHK-PEDES text-to-image | R@1: 44.02 -> 49.37 | Improved |
| add CMPC loss to CMPM (ResNet-152) (ablation) | CMPM · Flickr30K text-to-image | R@1: 35.7 -> 37.3 | Improved |
| add CMPC loss to CMPM (MobileNet) (ablation) | CMPM · MSCOCO 1K image-to-text | R@1: 51.4 -> 52.9 | Improved |
| replace MobileNet with ResNet-152 image encoder | CMPM · Flickr30K image-to-text | R@1: 37.1 -> 48.3 | Improved |
| replace KL(q||p) with CMPM (KL direction) at batch size 128 (ablation) | matching loss · CUHK-PEDES text-to-image | R@1: 36.06 -> 42.43 | Improved |
| use CMPM instead of N-pair loss for image-to-text at batch size 64 (ablation) | matching loss · CUHK-PEDES image-to-text | R@1: 12.07 -> 51.98 | Improved |
Closest papers by meaning
- Variational Adapter for Cross-modal Similarity Representation (ICML 2026 · similarity 0.88)
- CAMP: Cross-Modal Adaptive Message Passing for Text-Image Retrieval (ICCV 2019 · similarity 0.87)
- Look, Imagine and Match: Improving Textual-Visual Cross-Modal Retrieval With Generative Models (CVPR 2018 · similarity 0.87)
- Identity-Aware Textual-Visual Matching With Latent Co-Attention (ICCV 2017 · similarity 0.87)
- Cross-Modal and Uni-Modal Soft-Label Alignment for Image-Text Retrieval (AAAI 2024 · similarity 0.85)
- SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger (AAAI 2024 · similarity 0.84)
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.
