Research map / Video / Video-language understanding
Player-Centric Multimodal Prompt Generation for Large Language Model Based Identity-Aware Basketball Video Captioning
Zeyu Xi, Haoying Sun, Yaofei Wu, Junchi Yan, Haoran Zhang, Lifang Wu, Liang Wang, Changwen Chen
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 the player identification network (PIN) to the LLM captioner (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 66.5 -> 95.1 | Improved |
| add visual context learning module (VCLM) alone (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 66.5 -> 66.8 | No clear effect |
| add VCLM on top of PIN (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 95.1 -> 99.7 | Improved |
| add bidirectional semantic interaction module (BSIM) on top of PIN (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 95.1 -> 103.4 | Improved |
| add VCLM when PIN and BSIM are present (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 103.4 -> 105.3 | Improved |
| output both enhanced video and player features from BSIM instead of one (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 101.7 / 102.3 -> 105.3 | Improved |
| add multimodal identity prompt to LLM decoder (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 66.5 -> 105.3 | Improved |
| increase BSIM down-projection dimension from 512 to 768 (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | BLEU-4: 18.8 -> 18.0 | Hurt |
| reduce BSIM down-projection dimension from 512 to 128 (ablation) | LLM-IAVC (Llama3.2-3B) · NBA-Identity | CIDEr: 105.3 -> 99.6 | Hurt |
| use TimeSformer instead of ResNet-50 as player identification backbone (ablation) | Player identification network · NBA-Identity player classification | MCA: 90.12% -> 91.40% | Improved |
| use C3D or I3D instead of image backbones for player identification (ablation) | Player identification network · NBA-Identity player classification | MCA: C3D 50.86%, I3D 60.24% vs TimeSformer 91.40% | Hurt |
| use Llama3.2-3B decoder instead of Qwen2.5-3B or GPT-2 (ablation) | LLM-IAVC with prompt · NBA-Identity | CIDEr: 103.5 (Q-3B), 94.1 (GPT-2) -> 105.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- OmniViD — OmniViD: A Generative Framework for Universal Video Understanding (CVPR 2024) · better on 1 of 1 shared result
- SwinBERT — SwinBERT: End-to-End Transformers With Sparse Attention for Video Captioning (CVPR 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Universal Soccer Video Understanding (CVPR 2025 · similarity 0.85)
- BASKET: A Large-Scale Video Dataset for Fine-Grained Skill Estimation (CVPR 2025 · similarity 0.84)
- SportsHHI: A Dataset for Human-Human Interaction Detection in Sports Videos (CVPR 2024 · similarity 0.84)
- SportR: A Benchmark for Multimodal Large Language Model Reasoning in Sports (ICLR 2026 · similarity 0.84)
- FineSports: A Multi-person Hierarchical Sports Video Dataset for Fine-grained Action Understanding (CVPR 2024 · similarity 0.84)
- ViSTec: Video Modeling for Sports Technique Recognition and Tactical Analysis (AAAI 2024 · similarity 0.83)
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.
