Research map / Vision-language and multimodal / CLIP and zero-shot
Fast Zero-Shot Image Tagging
Read the paperCheck 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 a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · NUS-WIDE conventional tagging (81 tags) | MiAP: 52 -> 55 | Helped |
| use a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · NUS-WIDE zero-shot tagging (81 unseen tags) | MiAP: 40.1 -> 42.2 | Helped |
| use a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · NUS-WIDE tagging (4,093 unseen tags) | MiAP: 9.8 -> 8.5 | Hurt |
| use a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · IAPRTC-12 conventional tagging (291 tags) | MiAP: 46 -> 56 | Helped |
| use a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · IAPRTC-12 zero-shot tagging (58 unseen tags) | MiAP: 23.1 -> 20.3 | Hurt |
| use a neural network mapping instead of a linear mapping (ablation) | Fast0Tag · IAPRTC-12 seen/unseen tagging | MiAP: 42.9 -> 45.9 | Helped |
| normalize each image's pairwise ranking loss by its number of ranking constraints (ablation) | Fast0Tag neural network · NUS-WIDE validation | MiAP: about 2% | Helped |
| use RankNet instead of the other tested pairwise ranking losses (ablation) | Fast0Tag neural network · NUS-WIDE validation | MiAP: about 2% | Helped |
| modify LabelEM by removing objective terms involving duplicated images (ablation) | LabelEM · IAPRTC-12 zero-shot and seen/unseen tagging | MiAP: 11.5 -> 17.6; 13.8 -> 20.1 | Helped |
| use all positive-negative tag pairs across the batch instead of pairing only within each image (ablation) | Fast0Tag neural network · NUS-WIDE validation | MiAP | Helped |
| use the hinge loss instead of RankNet (ablation) | Fast0Tag neural network · NUS-WIDE validation | MiAP | Hurt |
Closest papers by meaning
- Tag2Text: Guiding Vision-Language Model via Image Tagging (ICLR 2024 · similarity 0.87)
- Mining Fine-Grained Image-Text Alignment for Zero-Shot Captioning via Text-Only Training (AAAI 2024 · similarity 0.87)
- Learning Transferable Visual Models From Natural Language Supervision (ICML 2021 · similarity 0.87)
- Zero-shot Referring Expression Comprehension via Structural Similarity Between Images and Captions (CVPR 2024 · similarity 0.85)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.85)
- Zero-Shot Image Captioning with Multi-type Entity Representations (AAAI 2025 · 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.