Research map / Architectures and efficiency / Pruning and efficient tuning
CLIPPING: Distilling CLIP-Based Models With a Student Base for Video-Language Retrieval
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Distill a CLIP-based video-language teacher into a MobileViT-v2 student using CLIPPING | Video-language retrieval · MSR-VTT 1k split, text-to-video | R@1: 38.9 -> 40.6 | Helped |
| Add temporal KD (ablation) | MobileViTv2 student · MSR-VTT 1k validation, text-to-video | R@1: 25.7 -> 28.8 | Helped |
| Add spatial KD alongside temporal KD (ablation) | MobileViTv2 student · MSR-VTT 1k validation, text-to-video | R@1: 28.8 -> 33.0 | Helped |
| Add SAB intermediate-layer KD alongside temporal and spatial KD (ablation) | MobileViTv2 student · MSR-VTT 1k validation, text-to-video | R@1: 33.0 -> 37.6 | Helped |
| Add global cross-modality KD (ablation) | CLIPPING with temporal, spatial, and SAB KD · MSR-VTT 1k validation, text-to-video | R@1: 37.6 -> 39.6 | Helped |
| Add local cross-modality KD (ablation) | CLIPPING with temporal, spatial, SAB, and global cross-modality KD · MSR-VTT 1k validation, text-to-video | R@1: 39.6 -> 40.7 | Helped |
| Replace OL2OL KD with SAB KD (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 34.6 -> 37.6 | Helped |
| Replace TAB KD with SAB KD (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 35.1 -> 37.6 | Helped |
| Use sequential masks in SAB KD (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 37.1 -> 37.6 | Helped |
| Remove the temporal Transformer (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 40.6 -> 38.6 | Hurt |
| Compress the text encoder with TinyBERT (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 40.6 -> 39.8 | Hurt |
| Pre-train the vision encoder on ImageNet21K (ablation) | CLIPPING · MSR-VTT 1k split, text-to-video | R@1: 40.6 -> 40.7 | No clear effect |
Closest papers by meaning
- CLIP-KD: An Empirical Study of CLIP Model Distillation (CVPR 2024 · similarity 0.91)
- Retaining Knowledge and Enhancing Long-Text Representations in CLIP through Dual-Teacher Distillation (CVPR 2025 · similarity 0.89)
- Meta CLIP 2: A Worldwide Scaling Recipe (NeurIPS 2025 · similarity 0.88)
- Building Vision-Language Models on Solid Foundations with Masked Distillation (CVPR 2024 · similarity 0.88)
- Fewer Steps, Better Performance: Efficient Cross-Modal Clip Trimming for Video Moment Retrieval Using Language (AAAI 2024 · similarity 0.87)
- Vgent: Graph-based Retrieval-Reasoning-Augmented Generation For Long Video Understanding (NeurIPS 2025 · similarity 0.86)
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.