Research map / Vision-language and multimodal / Human and embodied modeling
Lenses: Toward Polysemous Vision-Language Understanding
Hani Alomari, Ali Asgarov, Chris Thomas
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 lens-conditioned caption-to-slot alignment loss (ablation) | Lenses (BGE-VL-MLLM) · Lenses test set (All) | RSUM: 502.7 -> 510.6 | Improved |
| add intra-image prompt diversity loss (ablation) | Lenses (BGE-VL-MLLM) · Lenses test set (All) | RSUM: 510.6 -> 513.3 | Improved |
| fine-tune BGE-VL-MLLM on the Lenses dataset | BGE-VL-MLLM · Lenses test set (All) I->T | R@1: 70.67 -> 80.89 | Improved |
| add lens-aware similarity and training objectives on top of fine-tuning | BGE-VL-MLLM · Lenses test set (All) I->T | R@1: 80.89 -> 89.09 | Improved |
| add lens-aware training | BGE-VL-MLLM · Lenses test set (Figurative) I->T | R@1: 47.52 -> 56.02 | Improved |
| fine-tune on Lenses with lens-aware objectives | BGE-VL-MLLM · ArtEmis (Human Caption) I->T | R@1: 14.03 -> 15.67 | Improved |
| scale InternVL3.5 prompt generator from 4B to 14B (ablation) | Lenses · Lenses test set macro | R@1: 45.44 -> 44.77 | No clear effect |
| use InternVL3.5-38B instead of 14B (ablation) | Lenses · Lenses test set macro I->T | R@1: 44.77 -> 48.30 | Improved |
| train with images instead of text-only prompt-caption matching | Lenses · Lenses | R@1: 34.13 (P->C text-only) vs 80+ All R@1 | Improved |
| use Lenses lens-aware model over fine-tuned BGE-VL-MLLM for multi-lens coverage | Lenses · Lenses multi-lens setup | All@10: 21.4 -> 36.8 | Improved |
Closest papers by meaning
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- LensVLM: Selective Context Expansion for Compressed Visual Representation of Text (NeurIPS 2026 · similarity 0.87)
- Natural Language Inference Improves Compositionality in Vision-Language Models (ICLR 2025 · similarity 0.87)
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.87)
- Seeing the Abstract: Translating the Abstract Language for Vision Language Models (CVPR 2025 · similarity 0.87)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · 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.
