All operation is based KTH dataset, including Alexnet, Vgg16, LSTM, CNN-LSTM.
- Data reading: allreading, batch reading, tfrecord reding, list reading, tfrecord test
- Data preprocess: shuffle, resize, crop, standardization
- Model save and restore to test
- Tensorboard: loss, accuracy, feturemap show
- Conv and fc layer feature extractor then sabved mat file
- Dropout out, batch normalization, optimizer select, learning rate select
- Finetune base imagenent pretrained parmaeter
- Model convert to pb
- Extracted feature classfiy with SVM