Research map / Video / Video-language understanding
Retrospective Encoders for Video Summarization
Ke Zhang, Kristen Grauman, Fei Sha
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 retrospective-encoder losses (l_match and l_mismatch) to re-seq2seq (ablation) | re-seq2seq · SumMe | F-score: 43.2 -> 44.9 | Improved |
| add retrospective-encoder losses (ablation) | re-seq2seq · TVSum | F-score: 61.9 -> 63.9 | Improved |
| add retrospective-encoder losses (ablation) | re-seq2seq · VTW | F-score: 45.1 -> 48.0 | Improved |
| add l_match only (lambda>0, eta=0) (ablation) | re-seq2seq · VTW | F-score: 45.1 -> 47.7 | Improved |
| add l_mismatch only (lambda=0, eta>0) (ablation) | re-seq2seq · TVSum | F-score: 61.9 -> 63.2 | Improved |
| add l_mismatch in addition to l_match (ablation) | re-seq2seq · TVSum | F-score: 63.0 -> 63.9 | Improved |
| use hierarchical frame-shot modeling instead of frame-only seq2seq | re-seq2seq (lambda=0, eta=0) · TVSum | F-score: 56.3 -> 61.9 | Improved |
| replace cross-entropy summary loss of H-RNN with regression loss | re-seq2seq (lambda=0, eta=0) · SumMe | F-score: 43.6 -> 43.2 | Hurt |
| learn shot boundaries with LSTM instead of KTS (ablation) | re-seq2seq · TVSum | F-score: 63.6 -> 63.9 | Improved |
| learn shot boundaries with LSTM instead of KTS (ablation) | re-seq2seq · SumMe | F-score: 44.8 -> 44.9 | Improved |
| use test data in unlabeled losses (transductive setting) | re-seq2seq · TVSum | F-score: 63.9 -> 65.4 | Improved |
| add 1800 unlabeled videos by joint-training | re-seq2seq · TVSum | F-score: 63.9 -> 65.2 | Improved |
Closest papers by meaning
- Video Summarization by Learning From Unpaired Data (CVPR 2019 · similarity 0.90)
- Weakly-supervised Video Summarization using Variational Encoder-Decoder and Web Prior (ECCV 2018 · similarity 0.90)
- Video Summarization with Large Language Models (CVPR 2025 · similarity 0.88)
- Scaling Up Video Summarization Pretraining with Large Language Models (CVPR 2024 · similarity 0.87)
- Video Summarization Using Denoising Diffusion Probabilistic Model (AAAI 2025 · similarity 0.87)
- Seq2Time: Sequential Knowledge Transfer for Video LLM Temporal Grounding (CVPR 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.
