BoxComm-Benchmark

BoxComm-Benchmark is the official benchmark release for standardized BoxComm evaluation.

Resources

• Project Page: https://gouba2333.github.io/BoxComm
• Paper: http://arxiv.org/abs/2604.04419
• Code: https://github.com/gouba2333/BoxComm

Included files

• manifests/benchmark_manifest_eval_v1.jsonl
• metadata/eval_metadata_v1.csv
• examples/generation_prediction_example.jsonl
• examples/streaming_prediction_example.jsonl
• metrics/version.txt

Benchmark tasks

1. Category-conditioned commentary generation
2. Commentary rhythm and class-distribution evaluation

Evaluation scripts

Use the official scripts in the code repository:

• scripts/eval_metrics.py
• scripts/eval_streaming_cls_metrics.py

Prediction formats

Generation

{"video_id": 478, "segment_index": 0, "t_mid": 1.0, "pred_text": "..."}

Streaming

{"video_id": 478, "responses": [{"start_time": 0.0, "end_time": 1.0, "response": "..."}, ...]}

Citation

@article{wang2026boxcomm,
  title={BoxComm: Benchmarking Category-Aware Commentary Generation and Narration Rhythm in Boxing},
  author={Wang, Kaiwen and Zheng, Kaili and Deng, Rongrong and Shi, Yiming and Guo, Chenyi and Wu, Ji},
  journal={arXiv preprint arXiv:2604.04419},
  year={2026}
}