File size: 2,062 Bytes
6fb5d57
9685f7b
6fb5d57
 
 
 
 
 
 
70f5f26
1d78f26
70f5f26
 
 
1d78f26
70f5f26
 
 
1d78f26
70f5f26
 
 
 
1d78f26
 
 
 
70f5f26
 
1d78f26
70f5f26
 
 
 
eafc2c8
 
 
 
70f5f26
eafc2c8
 
70f5f26
 
1d78f26
 
 
70f5f26
 
 
 
 
 
 
 
 
 
1d78f26
 
70f5f26
 
 
 
 
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
---
title: Submission Template
emoji: 🔥
colorFrom: yellow
colorTo: green
sdk: docker
pinned: false
---


# Object Detector for forest fire smoke

## Model Description

This is a frugal object detector use to detect fire smoke, as part of the Frugal AI Challenge 2024. It is based of the yolo model series

### Intended Use

- **Primary intended uses**: Detect fire smoke on photos of forests, in different natural settings
- **Primary intended users**: Researchers and developers participating in the Frugal AI Challenge

## Training Data

The model uses the pyronear/pyro-sdis dataset:
- Size: ~33 600 examples
- Split: 88% train, 12% test
- Images with smoke or no smoke

### Labels
Smoke

## Performance

### Metrics
All reported on the test set
- **Accuracy**: ~ 90.8% 
- **Precision**: ~ 91.7%
- **Recall**: ~ 97.8%  
- **Environmental Impact**:
  - Emissions tracked in gCO2eq: 0.205
  - Energy consumption tracked in Wh: 3.66

### Model Architecture
Based of YOLOv11, see https://arxiv.org/abs/2410.17725, fine tuned on the pyronear dataset. The network is pruned and quantized to be as compressed as possible. 

Inference should ideally performed on GPU - the speed bump is drastic, it is more energy efficient than CPU inference which takes much longer.

## Environmental Impact

Environmental impact is tracked using CodeCarbon, measuring:
- Carbon emissions during inference
- Energy consumption during inference

This tracking helps establish a baseline for the environmental impact of model deployment and inference.

## Limitations
- Quantization was performed to FP16 - INT8 could compress even more but the accuracy drop was too big. Finding a way to smartly quantize and calibrate to INT8 could be interesting
- To maximize inference speed even more, the model can be converted to TensorRT - it is note done in this repository, as the same type of GPU needs to be used both for exporting to TensorRT and inferencing with TensorRT

## Ethical Considerations

- Environmental impact is tracked to promote awareness of AI's carbon footprint
```