Research map / Label-efficient and robust learning / Self-supervised and contrastive
SLIP: Self-Supervision Meets Language-Image Pre-training
Norman Mu, Alexander Kirillov, David Wagner, Saining Xie
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 SimCLR self-supervision to CLIP | CLIP ViT-B/16 · ImageNet zero-shot | top-1 accuracy (%): 37.6 -> 42.8 | Improved |
| add SimCLR self-supervision to CLIP | CLIP ViT-S/16 · ImageNet zero-shot | top-1 accuracy (%): 32.7 -> 38.3 | Improved |
| add SimCLR self-supervision to CLIP | CLIP ViT-L/16 · ImageNet zero-shot | top-1 accuracy (%): 40.4 -> 46.2 | Improved |
| add CLIP language supervision to SimCLR | SimCLR ViT-B/16 · ImageNet linear | top-1 accuracy (%): 64.0 -> 72.1 | Improved |
| add CLIP language supervision to SimCLR: finetuning | SimCLR ViT-B/16 · ImageNet finetuning | top-1 accuracy (%): 82.5 -> 82.6 | No clear effect |
| pre-train on YFCC15M instead of ImageNet (ablation) | SimCLR ViT-B/16 · ImageNet linear | top-1 accuracy (%): 74.5 -> 64.0 | Hurt |
| train for 100 epochs instead of 25 (ablation) | SLIP ViT-B/16 · ImageNet zero-shot | top-1 accuracy (%): 42.8 -> 45.0 | Improved |
| scale model from ViT-B to ViT-L (ablation) | SLIP · ImageNet zero-shot | top-1 accuracy (%): 45.0 -> 47.9 | Improved |
| use multi-task training instead of SimCLR then CLIP finetuning (ablation) | SLIP · ImageNet zero-shot | top-1 accuracy (%): 41.1 -> 42.8 | Improved |
| add color + blur augmentation to CLIP (ablation) | CLIP ViT-B/16 · ImageNet zero-shot | top-1 accuracy (%): 37.6 -> 38.4 | Improved |
| add resize crop + flip augmentation to CLIP (ablation) | CLIP ViT-B/16 · ImageNet zero-shot | top-1 accuracy (%): 37.6 -> 36.0 | Hurt |
| decouple self-supervision and text-supervision (ablation) | SLIP ViT-B/16 · ImageNet zero-shot | top-1 accuracy (%): 42.8 -> 42.7 | No clear effect |
Closest papers by meaning
- Demystifying CLIP Data (ICLR 2024 · similarity 0.88)
- Diffusion Feedback Helps CLIP See Better (ICLR 2025 · similarity 0.88)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.87)
- SuperCLIP: CLIP with Simple Classification Supervision (NeurIPS 2025 · similarity 0.87)
- Interpreting and Analysing CLIP's Zero-Shot Image Classification via Mutual Knowledge (NeurIPS 2024 · similarity 0.87)
- DINOv2 Meets Text: A Unified Framework for Image- and Pixel-Level Vision-Language Alignment (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.
