INTRODUCTION

Requirements:

Python 3.5

PyQt5 5.9.2

opencv-python 3.4.0.12

tensorflow-gpu 1.5.0

CUDA 8.0

easydict

Use:

1. run train_and_detection/win_entry.py

2. "open model" --> (summary/cnn3d_17)

3. "open video"

Model and TestVideo:

链接: https://pan.baidu.com/s/1O2h2lsNYtgtdDGYQI7aG3Q 密码: gzdq

Train:

run train_and_detection/train.py

Reference:

[1]Learning Spatio-Temporal Representation with Pseudo-3D Residual,ICCV2017

Spatio-Temporal Deep Neursl Network Based Video Smoke Detection

1.model

2Dto3D：

3D：

conv3d block(A,B,C)：

3D_DenseNet：

Result1：

Result2：

PS：

A simple example train or test 3DCNN：

A simple example as follows：
https://github.com/TianzhongSong/3D-ConvNets-for-Action-Recognition
Change /models/densenet_3d.py，and replace 3 * 3 * 3 Conv with 1 * 3 * 3 Conv and 3 * 3 * 1 Conv, modify in accordance with 3D_DenseNet，then you can train your own model.