Python 3.5
PyQt5 5.9.2
opencv-python 3.4.0.12
tensorflow-gpu 1.5.0
CUDA 8.0
easydict
-
run train_and_detection/win_entry.py
-
"open model" --> (summary/cnn3d_17)
-
"open video"
链接: https://pan.baidu.com/s/1O2h2lsNYtgtdDGYQI7aG3Q 密码: gzdq
run train_and_detection/train.py
[1]Learning Spatio-Temporal Representation with Pseudo-3D Residual,ICCV2017
2Dto3D：
3D：
conv3d block(A,B,C)：
3D_DenseNet：
Result1：
Result2：
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.