Research map / Vision-language and multimodal / Visual question answering
LingoQA: Video Question Answering for Autonomous Driving
Ana-Maria Marcu, Long Chen, Jan Hünermann, Alice Karnsund, Benoit Hanotte, Prajwal Chidananda, Saurabh Nair, Vijay Badrinarayanan, Alex Kendall, Jamie Shotton, Elahe Arani, Oleg Sinavski
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 LingoQA fine-tuning to generic VQA pre-trained model (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 33.60 -> 60.80 | Improved |
| add generic VQA pre-training (GQA, SVIT) before fine-tuning (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 56.60 -> 60.80 | Improved |
| combine Action and Scenery fine-tuning datasets (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 55.40 -> 60.80 | Improved |
| use 5 video frames instead of a single frame (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 57.00 -> 60.80 | Improved |
| increase frames from 5 to 7 (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 60.80 -> 60.60 | No clear effect |
| use late fusion instead of early fusion (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 48.40 -> 60.80 | Improved |
| use late fusion instead of mid fusion (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 59.20 -> 60.80 | Improved |
| use Vicuna-1.5-7B instead of Mistral-7B-Instruct (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 58.00 -> 60.80 | Improved |
| use Vicuna-1.5-7B instead of OPT-7B (ablation) | LingoQA Baseline · LingoQA evaluation | Lingo-Judge accuracy (%): 50.00 -> 60.80 | Improved |
| provide temporal context (5 frames instead of 1) to humans | Human evaluation · LingoQA evaluation | Lingo-Judge accuracy (%): 81.8 -> 96.6 | Improved |
| add visual input to fine-tuned LLM (LLaVA vs text-only Vicuna-7B) | fine-tuned Vicuna-7B · LingoQA evaluation | Lingo-Judge accuracy (%): 38.8 -> 59.0 | Improved |
| use Lingo-Judge instead of GPT-4 as automatic metric | evaluation metric · Correlation with human ratings | Spearman correlation: 0.941 -> 0.950 | Improved |
Closest papers by meaning
- NuScenes-QA: A Multi-Modal Visual Question Answering Benchmark for Autonomous Driving Scenario (AAAI 2024 · similarity 0.91)
- Open-ended VQA benchmarking of Vision-Language models by exploiting Classification datasets and their semantic hierarchy (ICLR 2024 · similarity 0.88)
- SimLingo: Vision-Only Closed-Loop Autonomous Driving with Language-Action Alignment (CVPR 2025 · similarity 0.88)
- Less Language, More Latents: Annotation-Efficient VLAs for Driving (NeurIPS 2026 · similarity 0.88)
- Embodied Scene Understanding for Vision Language Models via MetaVQA (CVPR 2025 · similarity 0.87)
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · similarity 0.87)
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.
