Research map / Video / Video-language understanding
EA-VTR: Event-Aware Video-Text Retrieval
Zongyang Ma, Ziqi Zhang, Yuxin Chen, Zhongang Qi, Chunfeng Yuan, Bing Li, Yingmin Luo, Xu LI, Xiaojuan Qi, Ying Shan, Weiming Hu
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 Event Content Learning (ECL) to video-level contrastive learning (ablation) | EA-VTR · MSRVTT zero-shot (1M pretrain) | R@1: 19.2 -> 21.5 | Improved |
| add Event Temporal Learning (ETL) to video-level contrastive learning (ablation) | EA-VTR · DiDeMo zero-shot (1M pretrain) | R@1: 20.3 -> 25.3 | Improved |
| combine ECL and ETL (ablation) | EA-VTR · DiDeMo zero-shot (1M pretrain) | R@1: 25.3 -> 27.3 | Improved |
| use Alternating Iteration Training instead of mixing original and synthesized pairs in a batch (ablation) | EA-VTR · DiDeMo zero-shot (1M pretrain) | R@1: 23.3 -> 27.3 | Improved |
| use event learning rather than only textual augmentation of captions (ablation) | EA-VTR · DiDeMo zero-shot (1M pretrain) | R@1: 23.8 -> 27.3 | Improved |
| use Top-p sampling instead of deterministic beam search for frame captions (ablation) | EA-VTR ECA · DiDeMo zero-shot (1M pretrain) | R@1: 23.9 -> 27.3 | Improved |
| replace BLIP captioner with stronger BLIP2 (ablation) | EA-VTR ECA · DiDeMo zero-shot (1M pretrain) | R@1: 27.3 -> 27.8 | No clear effect |
| synthesize videos by concatenation instead of weighted summation (ablation) | EA-VTR ETA · DiDeMo zero-shot (VTA+ETL) | R@1: 23.3 -> 25.3 | Improved |
| concatenate frame captions instead of web-crawled video captions for synthesized videos (ablation) | EA-VTR ETA · DiDeMo zero-shot (VTA+ETL) | R@1: 20.8 -> 25.3 | Improved |
| increase number of ECA augmented frames from 1 to 4 (ablation) | EA-VTR ECL · DiDeMo zero-shot (1M pretrain) | R@1: 23.6 -> 25.3 | Improved |
| use BERT instead of DistilBERT text encoder (ablation) | EA-VTR · DiDeMo zero-shot | R@1: 29.9 -> 32.7 | Improved |
| add event augmentation and event learning to a dual-encoder retriever | EA-VTR · ActivityNet Test of Time | time-order consistency (All): 50.3 -> 61.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Miles — MILES: Visual BERT Pre-training with Injected Language Semantics for Video-Text Retrieval (ECCV 2022) · better on 5 of 5 shared results
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
- MultiVENT 2.0: A Massive Multilingual Benchmark for Event-Centric Video Retrieval (CVPR 2025 · similarity 0.88)
- Bridging Information Asymmetry in Text-video Retrieval: A Data-centric Approach (ICLR 2025 · similarity 0.87)
- E.T. Bench: Towards Open-Ended Event-Level Video-Language Understanding (NeurIPS 2024 · similarity 0.87)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 2025 · similarity 0.87)
- Cross-modal Multi-task Learning for Multimedia Event Extraction (AAAI 2025 · similarity 0.86)
- EventLens: Event-Structure Reinforcement Learning for Video Understanding (NeurIPS 2026 · 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.
