Research map / Image generation / Text-to-image diffusion
Learning Deep Structure-Preserving Image-Text Embeddings
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 nonlinear embedding layers instead of using a linear embedding (ablation) | Image-text retrieval on Flickr30K · Flickr30K | R@1: 35.2 -> 40.3 | Helped |
| Use bi-directional rather than one-directional ranking constraints (ablation) | Nonlinear image-text embedding · Flickr30K sentence-to-image | R@1: 22.4 -> 28.1 | Helped |
| Add within-view structure-preserving constraints (ablation) | Nonlinear bi-directional image-text embedding · Flickr30K image-to-sentence | R@1: 39.3 -> 40.3 | Helped |
| Add within-view structure-preserving constraints (ablation) | Nonlinear bi-directional image-text embedding · MSCOCO 1000-image test | R@1: 47.5 -> 50.1 | Helped |
| Use mean word2vec features instead of Fisher vectors (ablation) | Nonlinear bi-directional embedding with structure · Flickr30K image-to-sentence | R@1: 40.3 -> 35.7 | Hurt |
| Use tf-idf features instead of Fisher vectors (ablation) | Nonlinear bi-directional embedding with structure · Flickr30K image-to-sentence | R@1: 40.3 -> 40.1 | No clear effect |
| Add hard-negative region mining and fine-tune the phrase-region embedding (ablation) | Flickr30K Entities phrase localization model (d) · Flickr30K Entities | R@1: 36.72 -> 43.89 | Helped |
| Add hard-negative region mining and fine-tune the phrase-region embedding (ablation) | Flickr30K Entities phrase localization model (d) · Flickr30K Entities | mAP(all): 41.13 -> 47.72 | Helped |
| Add only the phrase-side structure term (ablation) | Phrase-region embedding without negative mining · Flickr30K Entities | R@1: 35.83 -> 36.59 | Helped |
| Add both structure terms rather than neither (ablation) | Phrase-region embedding without negative mining · Flickr30K Entities | R@1: 35.83 -> 36.72 | Helped |
Closest papers by meaning
- Expressing Objects Just Like Words: Recurrent Visual Embedding for Image-Text Matching (AAAI 2020 · similarity 0.89)
- End-to-End Convolutional Semantic Embeddings (CVPR 2018 · similarity 0.88)
- Composing Object Relations and Attributes for Image-Text Matching (CVPR 2024 · similarity 0.86)
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.85)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.84)
- Text as Partial Constraint: Core–Residual Alignment for Robust Vision–Language Learning (NeurIPS 2026 · 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.