Research map / Vision-language and multimodal / Cross-modal learning
A Simple Baseline for Audio-Visual Scene-Aware Dialog
Check 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 |
|---|---|---|---|
| Use the question alone instead of the original multimodal baseline (ablation) | AVSD answer generation · AVSD v0.1 prototype validation | CIDEr: 0.766 -> 0.815 | Helped |
| Add spatial VGG video features to question and history (ablation) | q+h · AVSD v0.1 prototype validation | CIDEr: 0.843 -> 0.869 | Helped |
| Add audio features to question, history, and spatial VGG features (ablation) | q+h+vgg-spatial · AVSD v0.1 prototype validation | CIDEr: 0.869 -> 0.874 | Helped |
| Add attention to question, history, and spatial VGG features (ablation) | q+h+vgg-spatial · AVSD v0.1 prototype validation | CIDEr: 0.869 -> 0.908 | Helped |
| Include cross-data evidence in attention (ablation) | Proposed attention model · AVSD v0.1 prototype validation | CIDEr: 0.896 -> 0.941 | Helped |
| Include local evidence in attention (ablation) | Proposed attention model · AVSD v0.1 prototype validation | CIDEr: 0.917 -> 0.941 | Helped |
| Include the question prior in attention (ablation) | Proposed attention model · AVSD v0.1 prototype validation | CIDEr: 0.906 -> 0.941 | Helped |
| Share attention weights across video frames (ablation) | Proposed attention model · AVSD v0.1 prototype validation | CIDEr: 0.941 -> 0.923 | Hurt |
| Use spatial rather than temporal I3D video features (ablation) | AVSD video representation · AVSD v0.1 prototype validation | CIDEr: 0.886 -> 0.928 | Helped |
| Use 4 sampled VGG frames rather than 16 (ablation) | AVSD video representation · AVSD v0.1 prototype validation | CIDEr: 0.903 -> 0.941 | Helped |
| Use spatial VGG features rather than temporal I3D RGB features (ablation) | AVSD video representation · AVSD v0.1 prototype validation | CIDEr: 0.886 -> 0.919 | Helped |
| Use He initialization instead of default initialization (ablation) | AVSD model parameters · AVSD v0.1 prototype validation | CIDEr: 0.877 -> 0.913 | Helped |
Closest papers by meaning
- Crab: A Unified Audio-Visual Scene Understanding Model with Explicit Cooperation (CVPR 2025 · similarity 0.85)
- Towards Audio-Visual Navigation in Noisy Environments: A Large-Scale Benchmark Dataset and an Architecture Considering Multiple Sound-Sources (AAAI 2025 · similarity 0.85)
- RAO-Nav: Probing Omni-Language Models for Zero-shot Semantic Audio-Visual Navigation (NeurIPS 2026 · similarity 0.84)
- Hear, Localize, and Reason: Spatially Aware Scene Understanding for Audio-visual LLMs (NeurIPS 2026 · similarity 0.84)
- CAVEN: An Embodied Conversational Agent for Efficient Audio-Visual Navigation in Noisy Environments (AAAI 2024 · similarity 0.84)
- Structure-Aware Multimodal Sequential Learning for Visual Dialog (AAAI 2024 · similarity 0.84)
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.