Research map / Video / Video-language understanding
Reconstruct and Represent Video Contents for Captioning via Reinforcement Learning
Wei Zhang, Bairui Wang, Lin Ma, Wei 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 global reconstructor to SA-LSTM (ablation) | SA-LSTM (Inception-V4) · MSR-VTT (test) | CIDEr: 39.9 -> 41.7 | Improved |
| add local reconstructor to SA-LSTM (ablation) | SA-LSTM (Inception-V4) · MSR-VTT (test) | CIDEr: 39.9 -> 42.7 | Improved |
| add local reconstructor to SA-LSTM trained with RL (ablation) | SA-LSTM (RL) · MSR-VTT (test) | CIDEr: 46.1 -> 48.7 | Improved |
| add global reconstructor to SA-LSTM (ablation) | SA-LSTM (Inception-V4) · MSVD | CIDEr: 76.2 -> 79.7 | Improved |
| add local reconstructor to SA-LSTM (ablation) | SA-LSTM (Inception-V4) · MSVD | CIDEr: 76.2 -> 80.3 | Improved |
| add local reconstructor to S2VT (ablation) | S2VT (Inception-V4) · MSVD | CIDEr: 66.7 -> 69.8 | Improved |
| add local reconstructor to SA-LSTM (ablation) | SA-LSTM · ActivityNet (val) | CIDEr: 28.50 -> 35.52 | Improved |
| add self-critical RL fine-tuning to SA-LSTM (ablation) | SA-LSTM · MSR-VTT (test) | CIDEr: 39.9 -> 46.1 | Improved |
| combine global and local reconstructors (g+l) in RL setting (ablation) | RecNet(RL) · MSVD | CIDEr: 83.6 -> 85.9 | Improved |
| combine global and local reconstructors (g+l) in RL setting (ablation) | RecNet(RL) · MSR-VTT (test) | CIDEr: 48.7 -> 49.5 | Improved |
| replace learned reconstructor attention with transposed decoder attention matrix (ablation) | RecNet_local · MSR-VTT (test) | CIDEr: 42.7 -> 40.4 | Hurt |
| reduce MSVD training data to 50% with reconstructor (ablation) | RecNet_local · MSVD (50% training data) | CIDEr gain over SA-LSTM: +9.3% vs +5.4% on full data | Improved |
Closest papers by meaning
- Reconstruction Network for Video Captioning (CVPR 2018 · similarity 0.96)
- Learning Hierarchical Modular Networks for Video Captioning (TPAMI 2023 · similarity 0.89)
- SibNet: Sibling Convolutional Encoder for Video Captioning (TPAMI 2020 · similarity 0.87)
- Temporal Deformable Convolutional Encoder-Decoder Networks for Video Captioning (AAAI 2019 · similarity 0.87)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.86)
- Video ReCap: Recursive Captioning of Hour-Long Videos (CVPR 2024 · 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.
