Het3DCNN

Heterogeneous 3D convolutional network using tensorflow

1 Environments:

Tensorflow 1.12.1 Python3

2 Dataset

Data are saved in tfrecord format. In this project, input features are:

• grayscale image
• x direction optical flow
• y direction optical flow
• x direction mbh
• y direction mbh

Data includes only first 10 classes from UCF50. (sorry about the small dataset)

TFrecord files can be downloaded by Google Drive.

Links:

training data

https://drive.google.com/drive/folders/1hGSgvXnA2V8qjSoxtYPbKMQSUlDLNFoI?usp=sharing Full matrerials are in the progress of reorganizing.

3 References:

1. HetConv: Heterogeneous Kernel-Based Convolutions for Deep CNNs
2. 3D Convolutional Neural Networks for Human Action Recognition