Research map / Video / Video-language understanding
TeachText: CrossModal Generalized Distillation for Text-Video Retrieval
Ioana Croitoru, Simion-Vlad Bogolin, Marius Leordeanu, Hailin Jin, Andrew Zisserman, Samuel Albanie, Yang Liu
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 TeachText distillation from multiple text-embedding teachers | CE+ · DiDeMo | geometric mean R1-R5-R10: 35.8 -> 40.4 | Improved |
| add TeachText distillation | CE+ · ActivityNet | geometric mean R1-R5-R10: 39.7 -> 46.3 | Improved |
| add TeachText distillation | CE+ · MSVD | geometric mean R1-R5-R10: 46.5 -> 46.6 | No clear effect |
| add TeachText distillation | MoEE · DiDeMo | geometric mean R1-R5-R10: 33.2 -> 40.2 | Improved |
| add TeachText distillation | CE · MSVD | geometric mean R1-R5-R10: 42.3 -> 42.6 | No clear effect |
| add TeachText distillation | CE+ · MSR-VTT full | geometric mean R1-R5-R10: 29.2 -> 30.4 | Improved |
| use TeachText rather than concatenating or averaging multiple text embeddings at test time (ablation) | CE+ · MSR-VTT | R1-R5-R10 geometric mean | Improved |
| add more teachers beyond three (ablation) | TeachText with CE+ student · MSR-VTT | R1-R5-R10 geometric mean | No clear effect |
| restrict distillation to top-K teacher ranks (Rank 1 or Rank 10) instead of the whole batch (ablation) | TeachText · MSR-VTT | R1-R5-R10 geometric mean | Hurt |
| distill cross-modal similarity matrix instead of Relational or Pdist distillation (ablation) | TeachText · MSR-VTT | R1-R5-R10 geometric mean | Improved |
| use mean aggregation of teacher similarity matrices instead of min or max (ablation) | TeachText · MSR-VTT | R1-R5-R10 geometric mean | Improved |
| use Huber loss instead of L1 or L2 for distillation (ablation) | TeachText · MSR-VTT | R1-R5-R10 geometric mean | Improved |
Later papers that compared with this method
1 later accepted paper reports TeachText as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Long-Form Video-Language Pre-Training with Multimodal Temporal Contrastive Learning (NeurIPS 2022 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Frozen in Time: A Joint Video and Image Encoder for End-to-End Retrieval (ICCV 2021 · similarity 0.89)
- X-Pool: Cross-Modal Language-Video Attention for Text-Video Retrieval (CVPR 2022 · similarity 0.88)
- DGL: Dynamic Global-Local Prompt Tuning for Text-Video Retrieval (AAAI 2024 · similarity 0.88)
- TempMe: Video Temporal Token Merging for Efficient Text-Video Retrieval (ICLR 2025 · similarity 0.88)
- Stitching Segments and Sentences towards Generalization in Video-Text Pre-training (AAAI 2024 · similarity 0.87)
- MUSE: Mamba Is Efficient Multi-scale Learner for Text-video Retrieval (AAAI 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.
