Research map / Vision-language and multimodal / Cross-modal learning
Thinking Fast and Slow: Efficient Text-to-Visual Retrieval With Transformers
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 backward captioning to forward captioning (ablation) | VirTex cross-attention retrieval on COCO · Flickr30K val | F-R@1: 33.4 -> 38.1 | Helped |
| Use gradually upsampled high-level feature maps rather than the VirTex conv5 7x7 map (ablation) | Slow cross-attention model · Flickr30K val | F-R@1: 38.1 -> 42.2 | Helped |
| Increase the Slow input resolution from 224 to 384 (ablation) | Slow cross-attention model · Flickr30K val | F-R@1: 42.2 -> 44.8 | Helped |
| Add Slow-teacher distillation to the Fast NCE BoW dual encoder | Fast NCE BoW · Flickr30K val, COCO-trained | F-R@1: 27.2 -> 37.7 | Helped |
| Add Slow-teacher distillation to the Fast NCE BoW dual encoder | Fast NCE BoW · Flickr30K val, CC-trained | F-R@1: 32.4 -> 33.4 | Helped |
| Use Fast retrieval followed by Slow re-ranking of 10 candidates | Fast & Slow retrieval · Flickr30K val, COCO-trained | F-R@1: 44.8 -> 47.2 | Helped |
| Use Fast retrieval followed by Slow re-ranking of 10 candidates | Fast & Slow retrieval · COCO val, COCO-trained | C-R@1: 39.0 -> 40.5 | Helped |
| Omit distillation before re-ranking 10 candidates (ablation) | Fast & Slow retrieval · Flickr30K val, COCO-trained | F-R@1: 47.2 -> 44.0 | Hurt |
| Use bag-of-words rather than a Transformer text encoder in the distilled dual encoder (ablation) | Fast student · Flickr30K val, COCO-trained | F-R@1: 27.4 -> 35.6 | Helped |
| Set distillation weight α/τ² to 10.0 rather than 0.0 at τ=1.0 (ablation) | Fast NCE BoW distillation · Flickr30K val | F-R@1: 26.8 -> 19.1 | Hurt |
| Use Fast and Slow video retrieval with 14x14 features and K=50 | VATEX retrieval · VATEX | R@1: 45.9 -> 50.5 | Helped |
Closest papers by meaning
- Towards Efficient and Effective Text-to-Video Retrieval with Coarse-to-Fine Visual Representation Learning (AAAI 2024 · similarity 0.89)
- Not Another Text Benchmark: Putting the “Visual" Back in Visual Question Answering for Large Video Models (NeurIPS 2026 · similarity 0.88)
- How to Make Cross Encoder a Good Teacher for Efficient Image-Text Retrieval? (CVPR 2024 · similarity 0.88)
- TIGeR: Unifying Text-to-Image Generation and Retrieval with Large Multimodal Models (ICLR 2025 · similarity 0.88)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.88)
- DGL: Dynamic Global-Local Prompt Tuning for Text-Video Retrieval (AAAI 2024 · 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.