Research map / Video / Video-language understanding
Reconstruction Network for Video Captioning
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 |
|---|---|---|---|
| Add global video-feature reconstruction to SA-LSTM | SA-LSTM (Inception-V4) · MSR-VTT-10K test | BLEU-4: 36.3 -> 38.3 | Helped |
| Add local attention-based video-feature reconstruction to SA-LSTM | SA-LSTM (Inception-V4) · MSR-VTT-10K test | BLEU-4: 36.3 -> 39.1 | Helped |
| Replace global reconstruction with local attention-based reconstruction (ablation) | RecNet · MSR-VTT-10K test | BLEU-4: 38.3 -> 39.1 | Helped |
| Use temporal attention rather than mean pooling to aggregate frame features | SA-LSTM versus MP-LSTM with VGG19 · MSR-VTT-10K test | BLEU-4: 34.8 -> 35.6 | Helped |
| Use temporal attention rather than mean pooling to aggregate frame features | SA-LSTM versus MP-LSTM with VGG19 · MSR-VTT-10K test | METEOR: 24.7 -> 25.4 | Helped |
| Add global reconstruction to S2VT | S2VT (Inception-V4) · MSVD test | BLEU-4: 39.6 -> 42.9 | Helped |
| Add local reconstruction to S2VT | S2VT (Inception-V4) · MSVD test | BLEU-4: 39.6 -> 43.7 | Helped |
| Add global video-feature reconstruction to SA-LSTM | SA-LSTM (Inception-V4) · MSVD test | BLEU-4: 45.3 -> 51.1 | Helped |
| Add local video-feature reconstruction to SA-LSTM | SA-LSTM (Inception-V4) · MSVD test | BLEU-4: 45.3 -> 52.3 | Helped |
| Set reconstruction-loss weight above zero (ablation) | RecNet · MSR-VTT-10K | BLEU-4 | Helped |
| Set reconstruction-loss weight too large (ablation) | RecNet · MSR-VTT-10K | BLEU-4 | Hurt |
Closest papers by meaning
- Temporal Deformable Convolutional Encoder-Decoder Networks for Video Captioning (AAAI 2019 · similarity 0.88)
- Regularizing RNNs for Caption Generation by Reconstructing the Past With the Present (CVPR 2018 · similarity 0.88)
- Non-Autoregressive Coarse-to-Fine Video Captioning (AAAI 2021 · similarity 0.88)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.86)
- Fine-Grained Captioning of Long Videos through Scene Graph Consolidation (ICML 2025 · similarity 0.86)
- Video ReCap: Recursive Captioning of Hour-Long Videos (CVPR 2024 · similarity 0.85)
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.