Research map / Vision-language and multimodal / Cross-modal learning
Probabilistic Embeddings for Cross-Modal Retrieval
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 |
|---|---|---|---|
| Use probabilistic embeddings instead of mean-only deterministic embeddings (ablation) | PCME on CUB Captions · CUB Captions unseen 50-class test | R-Precision i2t: 24.70 -> 26.28 | Helped |
| Use probabilistic embeddings instead of mean-only deterministic embeddings (ablation) | PCME on CUB Captions · CUB Captions unseen 50-class test | R-Precision t2i: 25.64 -> 26.77 | Helped |
| Use probabilistic embeddings instead of mean-only deterministic embeddings (ablation) | PCME on MS-COCO · MS-COCO 5K test | PMRP i2t: 34.0 -> 34.1 | Helped |
| Use probabilistic embeddings instead of mean-only deterministic embeddings (ablation) | PCME on MS-COCO · MS-COCO 5K test | PMRP t2i: 34.3 -> 34.4 | Helped |
| Add local attention to both μ and σ heads (ablation) | PCME · CUB Captions test | R-Precision i2t: 25.60 -> 26.28 | Helped |
| Add local attention to both μ and σ heads (ablation) | PCME · CUB Captions test | R-Precision t2i: 25.85 -> 26.77 | Helped |
| Add sigmoid and layer normalization to the σ module (ablation) | PCME · CUB Captions test | R-Precision i2t: 26.28 -> 23.81 | Hurt |
| Add sigmoid and layer normalization to the σ module (ablation) | PCME · CUB Captions test | R-Precision t2i: 26.77 -> 24.58 | Hurt |
| Allow a full diagonal covariance instead of isotropic σ (ablation) | PCME · CUB Captions test | R-Precision i2t: 25.7 -> 26.3 | Helped |
| Allow a full diagonal covariance instead of isotropic σ (ablation) | PCME · CUB Captions test | R-Precision t2i: 26.0 -> 26.8 | Helped |
| Use KL divergence as the test-time distribution distance rather than mean-only similarity (ablation) | PCME · CUB Captions test | R-Precision i2t: 24.70 -> 21.99 | Hurt |
| Use KL divergence as the test-time distribution distance rather than mean-only similarity (ablation) | PCME · CUB Captions test | R-Precision t2i: 25.64 -> 20.92 | Hurt |
Closest papers by meaning
- Improved Probabilistic Image-Text Representations (ICLR 2024 · similarity 0.89)
- Adaptive Prompt-Based Semantic Embedding with Inspire Potential of Implicit Knowledge for Cross-Modal Retrieval (AAAI 2025 · similarity 0.89)
- Same Concept, Different Directions: Cross-Modal Feature Heterogeneity in Sparse Autoencoders (NeurIPS 2026 · similarity 0.89)
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.89)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.89)
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.