Hugging Face's logo

UEmmanuel5
/
ProFSAM-Fire-Detector

Object Detection
ultralytics
yolo
yolo11
fire-detection
computer-vision
realtime
Model card Files
xet
Community
ProFSAM-Fire-Detector
File size: 2,607 Bytes 
eaf6b10
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5b1a4e4
 
9c4bdec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
101
102
103
104
105
106
107
108
 
---
license: agpl-3.0
library_name: ultralytics
pipeline_tag: object-detection
tags:
- yolo
- yolo11
- fire-detection
- computer-vision
- realtime
---

# YOLOv11n Fire Detector (ProFSAM)

**Paper:** https://arxiv.org/abs/2510.21782  
**Code:** https://github.com/UEmmanuel5/ProFSAM  
**Weights:** `Fire_best.pt`  

## Intended use
Bounding-box detection of fire to prompt SAM2/MobileSAM/TinySAM in the ProFSAM pipeline.

## Training data
FASDD subset: classes **fire** and **neither_firenorsmoke** only. Total images used: **51,749** (12,550 fire, 39,199 neither_firenorsmoke).

## Training setup (summary)
PyTorch 2.0, CUDA 12.4, main GPU: GTX 1050 Ti (4 GB).  
Ultralytics YOLOv11n initialized then trained 100 epochs.

### Script used
```python
from ultralytics import YOLO
import torch
torch.backends.cudnn.benchmark = True

model = YOLO("path/to/yolo11n.pt")
train_results = model.train(
    data="path/to/FASDD_CV_Fire/data.yaml",
    epochs=100,
    imgsz=640,
    batch=16,
    optimizer="AdamW",
    lr0=1e-4,
    lrf=0.01,
    dropout=0.15,
    weight_decay=5e-4,
    device=0,
    val=False,
    save=True,
    plots=False
)
````

## Detector metrics (FASDD fire-only subset)

| P     | R     | mAP@0.5 | mAP@0.5:0.95 |
| ----- | ----- | ------- | ------------ |
| 0.799 | 0.697 | 0.797   | 0.520        |


## Test data
If you do not have test images, I placed 4 test images from the khan dataset to be used during your testing phase.
Test the model [here](https://huggingface.co/spaces/UEmmanuel5/profsam-fire-demo)


## ⚖️ License

* **Model weights** (`Fire_best.pt`): **AGPL-3.0** (Ultralytics-trained).
* **Space code** (this repo): **Apache-2.0**.

---

## 🙏 Acknowledgements

* Ultralytics YOLO11
* SAM / SAM2 ecosystem and community

---

## 📚 Citation

If this demo or the model is useful in your research, please cite:

**Manuscript**

```bibtex
@inproceedings{profsam2025,
  author    = {Emmanuel U. Ugwu and Xinming Zhang},
  title     = {Promptable Fire Segmentation: Unleashing SAM2's Potential for Real-Time Mobile Deployment with Strategic Bounding Box Guidance},
  booktitle = {ICIGP '26},
  year      = {2026},
  address   = {Wuhan, China},
  month     = jan,
  note      = {to appear}
}
```

**Model/Code**

```bibtex
@software{profsam2025,
  author    = {Ugwu, Emmanuel U. and Zhang, Xinming},
  title     = {Promptable Fire Segmentation: Unleashing SAM2’s Potential for Real-Time Mobile Deployment with Strategic Bounding Box Guidance},
  year      = {2025},
  doi       = {10.5281/zenodo.17340313},
  url       = {https://doi.org/10.5281/zenodo.17340313}
}
```