Research map / Video / Video-language understanding
Sports-QA: A Large-Scale Video Question Answering Benchmark for Complex and Professional Sports
Haopeng Li, Andong Deng, Jun Liu, Hossein Rahmani, Yulan Guo, Bernt Schiele, Mohammed Bennamoun, Qiuhong Ke
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 standard Transformer encoder with Auto-Focus Transformer (GloVe) (ablation) | VideoQA baseline · Sports-QA | accuracy: 58.0 -> 59.2 | Improved |
| replace standard Transformer encoder with Auto-Focus Transformer (BERT) (ablation) | VideoQA baseline · Sports-QA | accuracy: 57.9 -> 59.1 | Improved |
| AFT on counterfactual questions (GloVe) (ablation) | VideoQA baseline · Sports-QA counterfactual | accuracy: 59.9 -> 56.3 | Hurt |
| use multiple focal lengths 3 and 9 instead of single focal length 3 (ablation) | AFT · Sports-QA val | accuracy: 58.6 -> 59.3 | Improved |
| use local focal length 9 instead of global attention only (ablation) | AFT · Sports-QA val | accuracy: 58.2 -> 59.0 | Improved |
| add global focus 80 to focal 9 (ablation) | AFT · Sports-QA val | accuracy: 59.0 -> 59.2 | Improved |
| pretrain on four other sports before fine-tuning on basketball (ablation) | AFT · Sports-QA basketball | accuracy: 57.1 -> 58.3 | Improved |
| pretrain on four other sports before fine-tuning on FineGym (ablation) | AFT · Sports-QA FineGym | accuracy: 48.5 -> 50.3 | Improved |
| use BERT instead of GloVe embeddings in BlindQA (ablation) | BlindQA · Sports-QA | F1-score: 14.9 -> 15.8 | Improved |
| use CoT prompting in Chat-UniVi-7B (ablation) | Chat-UniVi-7B · Sports-QA | accuracy: 31.52 -> 30.81 | Hurt |
| use CoT prompting in Video-LLaVA-7B (ablation) | Video-LLaVA-7B · Sports-QA | accuracy: 30.33 -> 30.55 | No clear effect |
| fine-tune Qwen2.5-3B on Sports-QA (ablation) | Qwen2.5-3B · Sports-QA | accuracy: 27.77 -> 63.71 | Improved |
Closest papers by meaning
- SPORTU: A Comprehensive Sports Understanding Benchmark for Multimodal Large Language Models (ICLR 2025 · similarity 0.90)
- Towards Temporal Compositional Reasoning in Long-Form Sports Videos (ECCV 2026 · similarity 0.90)
- SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports (ICLR 2026 · similarity 0.88)
- ActionAtlas: A VideoQA Benchmark for Domain-specialized Action Recognition (NeurIPS 2024 · similarity 0.87)
- VideoEspresso: A Large-Scale Chain-of-Thought Dataset for Fine-Grained Video Reasoning via Core Frame Selection (CVPR 2025 · similarity 0.86)
- Towards Universal Soccer Video Understanding (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.
