README.md

Pretrained CNN Models on Smoke Dataset

Requirements:

• Python 2.7+
• Keras
• Tensoflow
• H5py

Training parameters:

• Batch size = 10
• 100 epochs
• RMSProp optimizer (learning rate = 0.001)
• Training and evaluation sets: http://staff.ustc.edu.cn/~yfn/vsd.html (Sets 1 to 4)

Paper:

@INPROCEEDINGS{8004998, 
author={A. Filonenko and L. Kurnianggoro and K. H. Jo}, 
booktitle={2017 10th International Conference on Human System Interactions (HSI)}, 
title={Comparative study of modern convolutional neural networks for smoke detection on image data}, 
year={2017}, 
volume={}, 
number={}, 
pages={64-68}, 
keywords={neural nets;object detection;AlexNet;CNN;ImageNet dataset;Inception-V3;Inception-V4;ResNet;VGG;Xception;image data;inception-based networks;modern convolutional neural networks;smoke detection task;Biological neural networks;Convolution;Graphics processing units;Hardware;Neurons;Training;Videos}, 
doi={10.1109/HSI.2017.8004998}, 
ISSN={}, 
month={July},}

Preprocessed Yuan's smoke dataset (5GB)

AlexNet:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

Inception-V3:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

Inception-V4:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

ResNet-50:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

VGG-16:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

VGG-19:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

Xception:

• Scenario 1
• Scenario 2
• Scenario 3
• Scenario 4

License

GPL 3