Research map / Vision-language and multimodal / Visual question answering
NAAQA: A Neural Architecture for Acoustic Question Answering
Jerome Abdelnour, Jean Rouat, Giampiero Salvi
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 |
|---|---|---|---|
| Replace VisualFiLM image feature extractor with audio 2D convolution extractor (ablation) | VisualFiLM · CLEAR2 test | overall accuracy (%): 62.3 -> 77.6 | Improved |
| Compute time convolution first instead of frequency convolution in interleaved extractor (ablation) | NAAQA-Interleaved · CLEAR2 test | overall accuracy (%): 67.2 -> 78.0 | Improved |
| Use parallel time and frequency extractor instead of interleaved time-first (ablation) | NAAQA · CLEAR2 test | overall accuracy (%): 78.0 -> 78.5 | Improved |
| Use parallel extractor instead of 2D convolution extractor (ablation) | NAAQA · CLEAR2 test | overall accuracy (%): 77.6 -> 78.5 | Improved |
| Reduce GRU units G from 4096 to 512 and optimize hyperparameters (ablation) | NAAQA-Parallel · CLEAR2 test | overall accuracy (%): 78.5 -> 79.5 | Improved |
| Add MALiMo controller (ablation) | NAAQA-ParallelOptimized · CLEAR2 test | overall accuracy (%): 79.5 -> 78.2 | Hurt |
| Add MALiMo controller (ablation) | NAAQA-InterleavedFreq · CLEAR2 test | overall accuracy (%): 67.2 -> 64.8 | Hurt |
| Add MALiMo controller (ablation) | VisualFiLM · CLEAR2 test | overall accuracy (%): 62.3 -> 63.6 | Improved |
| Add MALiMo controller (ablation) | NAAQA-ParallelOptimized · DAQA' test | overall accuracy (%): 60.4 -> 64.3 | Improved |
| Add MALiMo controller (ablation) | NAAQA-2DConv-Optimized · DAQA' test | overall accuracy (%): 58.3 -> 54.4 | Hurt |
| Use parallel extractor instead of 2D conv extractor (ablation) | NAAQA-Optimized · DAQA' test | overall accuracy (%): 58.3 -> 60.4 | Improved |
| Add Time coordinate map in Resblocks (ablation) | NAAQA-Parallel · CLEAR2 test, coordinate map placement (Parallel, initial config) | test accuracy (%): 63.8 -> 79.0 | Improved |
Closest papers by meaning
- Question-Aware Gaussian Experts for Audio-Visual Question Answering (CVPR 2025 · similarity 0.88)
- Look, Listen, and Answer: Overcoming Biases for Audio-Visual Question Answering (NeurIPS 2024 · similarity 0.88)
- Audio-Visual Adaptive Fusion Network for Question Answering Based on Contrastive Learning (AAAI 2025 · similarity 0.87)
- When Open-Vocabulary Visual Question Answering Meets Causal Adapter: Benchmark and Approach (AAAI 2025 · similarity 0.86)
- CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark (NeurIPS 2024 · similarity 0.86)
- FRAMES-VQA: Benchmarking Fine-Tuning Robustness across Multi-Modal Shifts in Visual Question Answering (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.
