Research map / Vision-language and multimodal / CLIP and zero-shot
CLIMP: Contrastive Language-Image Mamba Pretraining
Nimrod Shabtay, Itamar Zimerman, Eli Schwartz, RAJA GIRYES
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 ViT/LLaMA encoders with VMamba/Mamba encoders | CLIP · ImageNet-O zero-shot | Acc@1: 40.1 -> 49.8 (vs RoPE-ViT) | Improved |
| replace ViT/LLaMA encoders with VMamba/Mamba encoders | CLIP · ImageNet-1K zero-shot | Acc@1: 40.7 -> 44.0 (vs RoPE-ViT) | Improved |
| replace ViT with VMamba | CLIP · Flickr30k retrieval | text R@5: 82.7 -> 87.5 | Improved |
| replace ViT with VMamba | CLIP · MSCOCO retrieval | text R@5: 57.5 -> 61.9 | Improved |
| replace ViT with VMamba | CLIP · ImageNet-A / ImageNet-R | Acc@1: 16.3 -> 15.6 (A, vs RoPE-ViT); 47.8 -> 46.6 (R) | Hurt |
| replace ViT with VMamba (Mamba-2) | CLIP · CIFAR-100 / FER2013 | Acc@1: 54.5 -> 50.7; 29.1 -> 22.2 | Hurt |
| use Mamba-1 instead of Mamba-2 text/vision blocks (ablation) | CLIMP · average over 28 zero-shot benchmarks | Acc@1: 26.9 -> 29.6 | Improved |
| use Mamba-1 instead of Mamba-2 (ablation) | CLIMP · OpenCLIP 35 datasets | avg acc@1: 34.48 -> 36.06 | Improved |
| use Mamba-2 for average Acc@5 (ablation) | CLIMP · 28 zero-shot benchmarks | Acc@5: 59.0 vs 58.5 | No clear effect |
Closest papers by meaning
- Contrastive Localized Language-Image Pre-Training (ICML 2025 · similarity 0.90)
- FFF: Fixing Flawed Foundations in Contrastive Pre-Training Results in Very Strong Vision-Language Models (CVPR 2024 · similarity 0.90)
- MLIP: Efficient Multi-Perspective Language-Image Pretraining with Exhaustive Data Utilization (ICML 2024 · similarity 0.90)
- Post-pre-training for Modality Alignment in Vision-Language Foundation Models (CVPR 2025 · similarity 0.89)
- Retrieval-Enhanced Contrastive Vision-Text Models (ICLR 2024 · similarity 0.89)
- Rethinking Contrastive Loss in CLIP Post-training: A Complementary Framework with Frozen Text Encoder (NeurIPS 2026 · similarity 0.89)
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.
