Research map / Vision-language and multimodal / Human and embodied modeling
Visual Word2Vec (vis-w2v): Learning Visually Grounded Word Embeddings Using Abstract Scenes
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 |
|---|---|---|---|
| Ground word embeddings using abstract-scene visual clusters | w2v-coco · Common sense assertion classification | AP (%): 72.2 -> 74.8 | Helped |
| Ground word embeddings using abstract-scene visual clusters | w2v-wiki · Common sense assertion classification | AP (%): 68.4 -> 74.2 | Helped |
| Use separate embeddings for primary object, relation, and secondary object roles (ablation) | vis-w2v-coco · Common sense assertion classification | AP (%): 74.5 -> 74.8 | Helped |
| Add the test-time visual similarity function (ablation) | vis-w2v-coco (shared) · Common sense assertion classification | AP (%): 74.5 -> 74.2 | Hurt |
| Use a lower hidden size of 50 (ablation) | vis-w2v-coco (separate) · Common sense assertion classification | AP (%): 74.8 -> 75.4 | Helped |
| Ground embeddings with abstract-scene features rather than real-image features (ablation) | vis-w2v · Common sense assertion classification | AP (%): 73.7 -> 74.8 | Helped |
| Initialize with Ren et al. embeddings before abstract-scene training (ablation) | vis-w2v-coco (separate) · Common sense assertion classification | AP (%): 74.8 -> 75.2 | Helped |
| Train with visual grounding | w2v-coco · Visual paraphrasing | AP (%): 94.6 -> 95.3 | Helped |
| Train with visual grounding | w2v-wiki · Visual paraphrasing | AP (%): 94.4 -> 95.1 | Helped |
| Reduce context from full descriptions to 5-word windows (ablation) | vis-w2v · Visual paraphrasing | AP gain over text baseline: 0.7% -> 0.6% gain | Hurt |
| Reduce context from full descriptions to single words (ablation) | vis-w2v · Visual paraphrasing | AP gain over text baseline: 0.7% -> 0.1% gain | Hurt |
| Ground embeddings using abstract-scene visual clusters | w2v-coco · Text-based image retrieval | Recall@10 (%): 47.6 -> 49.5 | Helped |
Closest papers by meaning
- Bridging Vision and Language Spaces with Assignment Prediction (ICLR 2024 · similarity 0.88)
- GroundingME: Exposing the Visual Grounding Gap in MLLMs through Multi-Dimensional Evaluation (CVPR 2026 · similarity 0.88)
- Textural or Textual: How Vision-Language Models Read Text in Images (ICML 2025 · similarity 0.87)
- Improved Visual Grounding through Self-Consistent Explanations (CVPR 2024 · similarity 0.87)
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.87)
- Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Grounding (CVPR 2025 · similarity 0.87)
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.