File size: 2,492 Bytes
0217140
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
---
license: other
task_categories:
  - video-classification
  - image-classification
tags:
  - boxing
  - sports
  - action-recognition
  - punch-detection
  - computer-vision
  - qwen
  - vision-llm
  - fine-tuning
pretty_name: Boxing Punch Classification Video Dataset
size_categories:
  - 1K<n<10K
---

# Boxing Punch Classification Video Dataset

Short video clips of boxing punches, labeled for training vision language models.

## Dataset Description

This dataset contains 2278 short video clips extracted from Olympic boxing competition footage,
each labeled with one of 8 punch/action categories. The clips are designed for fine-tuning
vision LLMs like Qwen-VL for boxing action recognition.

### Source

Derived from the [Olympic Boxing Punch Classification Video Dataset](https://www.kaggle.com/datasets/piotrstefaskiue/olympic-boxing-punch-classification-video-dataset)
by Piotr Stefański et al. Original recordings from a 2021 boxing league competition.

## Labels

| Label | Count |
|-------|-------|
| head_left_hand | 807 |
| head_right_hand | 390 |
| miss_left_hand | 387 |
| miss_right_hand | 209 |
| block_left_hand | 190 |
| body_left_hand | 111 |
| block_right_hand | 96 |
| body_right_hand | 88 |

**Total clips:** 2278
**Total duration:** 1562.9 seconds

## Dataset Structure

Each entry contains:
- `file_name`: Path to the video clip (.mp4)
- `label`: Action category (English)
- `start_time` / `end_time`: Timestamps in the source video
- `duration`: Clip length in seconds
- `source_video`: Original video filename
- `fps`: Frames per second

## Usage

```python
from datasets import load_dataset

ds = load_dataset("NealBeans/BoxingDataset")
```

### For Qwen-VL Fine-tuning

```python
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor

model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")

# Each clip can be used as video input with its label as the target
```

## License

This dataset is derived from research data and is for non-commercial/academic use only.
See the original dataset for full license terms.

## Citation

If you use this dataset, please cite the original research:

```bibtex
@article{stefanski2024boxing,
  title={Boxing Punch Detection with Single Static Camera},
  author={Stefa{\n}ski, Piotr and Kozak, Jan and Jach, Tomasz},
  journal={Entropy},
  volume={26},
  number={8},
  pages={617},
  year={2024}
}
```