Research map / Vision-language and multimodal / Human and embodied modeling
WikiCLIP: An Efficient Contrastive Baseline for Open-domain Visual Entity Recognition
Shan Ning, Longtian Qiu, Jiaxuan Sun, Xuming He
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 Wikipedia text descriptions to image-based entity representation (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 44.8 -> 59.7 | Improved |
| add entity images to text-only entity representation (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 50.8 -> 59.7 | Improved |
| add visual-based clustering alone as hard negative batching (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 59.7 -> 59.7 | No clear effect |
| add synthetic hard negatives alone without visual clustering (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 59.7 -> 58.9 | Hurt |
| combine visual clustering with synthetic hard negative generation (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 59.7 -> 61.2 | Improved |
| use EVA-CLIP 8B instead of CLIP ViT-L as visual encoder (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 41.6 -> 61.2 | Improved |
| use LLaMa3.2 1B instead of EVA-CLIP text encoder (ablation) | WikiCLIP-S · InfoSeek (val) | Overall accuracy: 58.1 -> 61.2 | Improved |
| increase encoded Wikipedia text length beyond 256 tokens (ablation) | WikiCLIP-S · InfoSeek (val) | accuracy | Hurt |
| scale LLM from 3B to 8B (ablation) | WikiCLIP · InfoSeek (val) | accuracy | No clear effect |
| scale LLM from 1B to 3B (ablation) | WikiCLIP · InfoSeek (val) | Unseen accuracy: 58.5 -> 60.3 | Improved |
| train with only 700 seen entities instead of full 7,943 (ablation) | WikiCLIP · OVEN unseen | Unseen accuracy: 58% -> 56% | No clear effect |
| use WikiCLIP-L instead of AutoVER 13B | OVEN entity recognition · OVEN Entity (val) | Unseen accuracy: 24.5 -> 28.5 | Improved |
Closest papers by meaning
- Web-Scale Visual Entity Recognition: An LLM-Driven Data Approach (NeurIPS 2024 · similarity 0.88)
- A Generative Approach for Wikipedia-Scale Visual Entity Recognition (CVPR 2024 · similarity 0.88)
- EvdCLIP: Improving Vision-Language Retrieval with Entity Visual Descriptions from Large Language Models (AAAI 2025 · similarity 0.87)
- MCL-NER: Cross-Lingual Named Entity Recognition via Multi-View Contrastive Learning (AAAI 2024 · similarity 0.86)
- Reverse Region-to-Entity Annotation for Pixel-Level Visual Entity Linking (AAAI 2025 · similarity 0.86)
- MOFI: Learning Image Representations from Noisy Entity Annotated Images (ICLR 2024 · similarity 0.83)
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.
