Research map / Video / Video-language understanding
Adaptive Feature Abstraction for Translating Video to Text
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 adaptive abstraction-level attention over intermediate C3D layers to spatiotemporal attention (ablation) | C3D video captioning with LSTM · YouTube2Text | BLEU-4: 48.18 -> 51.74 | Helped |
| Add adaptive abstraction-level attention over intermediate C3D layers to spatiotemporal attention (ablation) | C3D video captioning with LSTM · M-VAD | METEOR: 6.89 -> 7.72 | Helped |
| Add adaptive abstraction-level attention over intermediate C3D layers to spatiotemporal attention (ablation) | C3D video captioning with LSTM · MSR-VTT | BLEU-4: 42.04 -> 43.72 | Helped |
| Use convolutional transformations instead of MLP-based multi-level feature alignment (ablation) | dualAFR · YouTube2Text | BLEU-4: 35.99 -> 51.74 | Helped |
| Use convolutional transformations instead of max pooling plus MLP for multi-level feature alignment (ablation) | dualAFR · YouTube2Text | BLEU-4: 45.35 -> 51.74 | Helped |
| Use convolutional transformations instead of average pooling plus MLP for multi-level feature alignment (ablation) | dualAFR · YouTube2Text | BLEU-4: 48.53 -> 51.74 | Helped |
| Use convolutional transformations instead of hypercolumn alignment (ablation) | dualAFR · YouTube2Text | BLEU-4: 44.92 -> 51.74 | Helped |
| Use soft attention rather than hard attention (ablation) | dualAFR single model · M-VAD | METEOR: 7.12 -> 7.72 | Helped |
| Use hard attention rather than soft attention (ablation) | dualAFR ensemble · M-VAD | CIDEr: 9.03 -> 9.14 | Helped |
| Use an ensemble of 10 rather than a single model (ablation) | dualAFR soft attention · YouTube2Text | BLEU-4: 51.74 -> 53.94 | Helped |
| Use an ensemble rather than a single model (ablation) | dualAFR soft attention · MSR-VTT | BLEU-4: 43.72 -> 44.99 | Helped |
| Use adaptive intermediate-layer features and attention instead of only C3D fc7 features (ablation) | C3D video captioning with LSTM · YouTube2Text | BLEU-4: 45.34 -> 51.74 | Helped |
Closest papers by meaning
- Spatio-Temporal Dynamics and Semantic Attribute Enriched Visual Encoding for Video Captioning (CVPR 2019 · similarity 0.88)
- SwinBERT: End-to-End Transformers With Sparse Attention for Video Captioning (CVPR 2022 · similarity 0.88)
- Fully Convolutional Video Captioning with Coarse-to-Fine and Inherited Attention (AAAI 2019 · similarity 0.87)
- Distilling Vision-Language Models on Millions of Videos (CVPR 2024 · similarity 0.87)
- Dense Video Object Captioning from Disjoint Supervision (ICLR 2025 · similarity 0.85)
- CALVIN: Improved Contextual Video Captioning via Instruction Tuning (NeurIPS 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.