Research map / Video / Video-language understanding
Teaching Machines to Understand Baseball Games: Large-Scale Baseball Video Database for Multiple Video Understanding Tasks
Minho Shim, Young Hwi Kim, Kyungmin Kim, Seon Joo Kim
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 |
|---|---|---|---|
| increase GRU max sequence length from 64 to 256 (ablation) | CNN+GRU · BBDB action recognition | mAP (%): 36.2 -> 62.8 | Improved |
| add recurrent GRU over frames instead of I3D | I3D · BBDB action recognition | mAP (%): 44.2 -> 62.8 | Improved |
| add optical flow to RGB (two-stream) | Single frame · BBDB action recognition | mAP (%): 35.0 -> 42.3 | Improved |
| add oversampling (ablation) | CNN+GRU · BBDB imbalance handling (l = 256) | mAP (%): 67.0 -> 62.8 | Hurt |
| add hierarchical classification (ablation) | CNN+GRU · BBDB imbalance handling (l = 256) | mAP (%): 67.0 -> 50.4 | Hurt |
| add class weight adjustment (ablation) | CNN+GRU · BBDB imbalance handling (l = 256) | mAP (%): 67.0 -> 55.3 | Hurt |
| add focal loss (gamma = 2) on top of oversampling (ablation) | CNN+GRU · BBDB imbalance handling (l = 256) | mAP (%): 62.8 -> 61.6 | Mixed |
| change focal loss gamma to 0.5, 1 or 5 (ablation) | CNN+GRU+FL · BBDB imbalance handling (l = 256) | mAP (%) | No clear effect |
| mask the scoreboard during training (ablation) | CNN+GRU · BBDB ball event recognition | accuracy: 0.908 -> 0.719 | Hurt |
| mask the scoreboard during training (ablation) | CNN+GRU · BBDB strike event recognition | accuracy: 0.720 -> 0.394 | Hurt |
| use CDC instead of single frame | per-frame model · BBDB per-frame labeling | mAP (%): 9.25 -> 23.3 | Improved |
| use CDC instead of single frame | localization model · BBDB temporal localization (IoU 0.5) | mAP (%): 3.44 -> 11.3 | Improved |
Closest papers by meaning
- BASKET: A Large-Scale Video Dataset for Fine-Grained Skill Estimation (CVPR 2025 · similarity 0.83)
- FineSports: A Multi-person Hierarchical Sports Video Dataset for Fine-grained Action Understanding (CVPR 2024 · similarity 0.83)
- Towards Universal Soccer Video Understanding (CVPR 2025 · similarity 0.82)
- SPORTU: A Comprehensive Sports Understanding Benchmark for Multimodal Large Language Models (ICLR 2025 · similarity 0.82)
- SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos (CVPR 2024 · similarity 0.82)
- SoccerMaster: A Vision Foundation Model for Soccer Understanding (CVPR 2026 · similarity 0.81)
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.
