Research map / Vision-language and multimodal / CLIP and zero-shot
Controlling Embedding Spaces with Text-Conditioned Transformations
Joseph Fioresi, Fabian Caba, Pankaj Nathani, Shah Mubarak, Kushal Kafle
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 |
|---|---|---|---|
| use affine transform instead of translation-only (ablation) | text-conditioned hypernetwork · Clevr-4 query retrieval | mAP: 35.4 -> 77.6 | Improved |
| use affine transform instead of translation-only (ablation) | text-conditioned hypernetwork · Stanford-40 query retrieval | mAP: 43.8 -> 84.4 | Improved |
| use affine transform instead of translation-only (ablation) | text-conditioned hypernetwork · Stanford-40 cluster accuracy | accuracy: 55.6 -> 79.4 | Improved |
| add multimodal guidance loss (ablation) | hypernetwork · text-based classification ShotBench | accuracy: 31.3 -> 51.6 | Improved |
| add multimodal guidance loss (ablation) | hypernetwork · Stanford40 attribute-aware retrieval | mAP: 87.3 -> 91.2 | Improved |
| add multimodal guidance loss (ablation) | hypernetwork · ShotBench multi-clustering | clustering accuracy: 53.2 -> 51.6 | Hurt |
| blend action and location projected queries (alpha=0.5) | query transformation · Stanford40 joint-label retrieval | mAP (Both): 70.0 -> 80.9 | Improved |
| train with one color held out and test on it | hypernetwork · Clevr4 color retrieval | mAP: 22.9 -> 58.0 | Improved |
| raise multimodal loss weight from 0.1 to 1 (ablation) | hypernetwork · Clevr-4 retrieval and clustering | mAP / clustering accuracy: retrieval 73.4 -> 73.0, clustering 79.6 -> 81.5 | Mixed |
| remove multimodal loss (omega 0) versus 0.1 (ablation) | hypernetwork · Clevr-4 retrieval | mAP: 72.3 -> 73.4 | Improved |
| apply color-aware query transform on top of SEARLE-XL reranking | SEARLE-XL · FashionIQ average | R@10: 23.5 -> 24.6 | Improved |
| apply learned transformation to text embeddings with sentiment conditioning | E5-large-v2 · SST-5 | zero-shot accuracy: 29.6 -> 54.3 | Improved |
Closest papers by meaning
- Highlighting What Matters: Promptable Embeddings for Attribute-Focused Image Retrieval (NeurIPS 2025 · similarity 0.89)
- Disentangling Visual Embeddings for Attributes and Objects (CVPR 2022 · similarity 0.87)
- Is CLIP ideal? No. Can we fix it? Yes! (ICCV 2025 · similarity 0.87)
- Learning Attribute-Driven Disentangled Representations for Interactive Fashion Retrieval (ICCV 2021 · similarity 0.87)
- Continuous, Subject-Specific Attribute Control in T2I Models by Identifying Semantic Directions (CVPR 2025 · similarity 0.86)
- Retrieval-Enhanced Contrastive Vision-Text Models (ICLR 2024 · similarity 0.85)
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.
