Research map / Vision-language and multimodal / Cross-modal learning
IsoCLIP: Decomposing CLIP Projectors for Efficient Intra-modal Alignment
Simone Magistri, Dipam Goswami, Marco Mistretta, Bartłomiej Twardowski, Joost van de Weijer, Andrew D. Bagdanov
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 |
|---|---|---|---|
| replace standard projector with IsoCLIP isotropic-subspace projector | CLIP ViT-B/16 · Image-to-image retrieval | mAP avg: 46.3 -> 52.8 | Improved |
| replace standard projector with IsoCLIP | CLIP ViT-L/14 · Image-to-image retrieval | mAP avg: 54.3 -> 60.7 | Improved |
| replace standard projector with IsoCLIP | PE-Core-B-16 · Image-to-image retrieval | mAP avg: 58.6 -> 62.8 | Improved |
| replace OTI inversion with IsoCLIP | CLIP ViT-B/16 · Image-to-image retrieval | mAP avg: 49.4 -> 52.8 | Improved |
| replace OTI with IsoCLIP | CLIP ViT-B/16-open · Image-to-image retrieval | mAP avg: 58.7 -> 58.6 | No clear effect |
| replace OTI with IsoCLIP | CLIP ViT-B/32 · Image-to-image retrieval | query latency (ms): 1879 -> 7 | Improved |
| replace standard text projector with IsoCLIP | CLIP ViT-L/14 · Text-to-text retrieval | mAP avg: 38.2 -> 43.0 | Improved |
| replace OVI with IsoCLIP | CLIP ViT-B/16-open · Text-to-text retrieval | mAP avg: 45.3 -> 45.3 | No clear effect |
| replace OVI with IsoCLIP | CLIP ViT-B/32 · Text-to-text retrieval | query latency (ms): 11549 -> 6 | Improved |
| replace image-image NCM with IsoCLIP NCM | CLIP ViT-B/16 · Image classification | accuracy avg: 77.4 -> 80.1 | Improved |
| replace pre-projection features with IsoCLIP projector (ablation) (ablation) | CLIP ViT-B/16 · Image-to-image retrieval | mAP avg: 48.4 -> 52.8 | Improved |
| replace whitened projector with IsoCLIP projector (ablation) (ablation) | CLIP ViT-B/16 · Image-to-image retrieval | mAP avg: 49.0 -> 52.8 | Improved |
Closest papers by meaning
- Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion (ICLR 2025 · similarity 0.94)
- SoftCLIP: Softer Cross-Modal Alignment Makes CLIP Stronger (AAAI 2024 · similarity 0.90)
- SmartCLIP: Modular Vision-language Alignment with Identification Guarantees (CVPR 2025 · similarity 0.90)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.89)
- Do Vision and Language Encoders Represent the World Similarly? (CVPR 2024 · similarity 0.88)
- FineLIP: Extending CLIP's Reach via Fine-Grained Alignment with Longer Text Inputs (CVPR 2025 · similarity 0.88)
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.
