Setting and environment Both Windows and Linux:
Conda environment with Python 3.9:
pandas
OpenCV (Windows)
tensorflow (Windows)
tensorflow-gpu (OzStar)
tensorflow hub
sklearn
matplotlib
OzStar environment:
Source activate conda environment
module load anaconda3/5.1.0
GPU modules:
module load cudnn/8.1.0-cuda-11.2.0
module load cuda/11.2.0
SMART method: Data_processing.py SingleSelector.py GlobalSelector.py New-SingleSelector.ipynb New-GlobalSelector.ipynb New-SelectorModel.ipynb
CNNs:
Preprocessing_UCF101.py TimeDistributed_ResNet50_MLP.py TimeDistributed_ResNet50_LSTM_MLP.py
Backend model: ResNet50+MLP after Selector model.ipynb
Process pipeline:
We need to perform “New-SingleSelector.ipynb” and “New-GlobalSelector.ipynb” training and get the trained weight. After that, we use “New-SelectorModel.ipynb” to complete the SMART method. After that, we use the SMART method output into “Preprocessing_UCF101.py”, and it will create a new folder. Finally, use “ResNet50+MLP after Selector model” to get the final results.
All processes are based on Python only.