Deep CNN and RNN

The classic deep convolution and recurrent neural network project with TensorFlow. You can learn and run CNN easily.

• Train/test/inference images included.
• Visualize images with matplotlib.
• Support checkpoint and tensorboard.
• Easy to extend more convolution layers.
• Support CNN, LSTM, Bicondictional LSTM and Stacked LSTM.

Train

./pokemon_classifer.py --epoch_number 100

Inference

./pokemon_classifer.py --mode inference --image ./data/inference/Pikachu.png

Export model

./pokemon_classifer.py --epoch_number 0

Run TensorFlow serving

./tensorflow_model_server --port=9000 --model_name=deep_cnn --model_base_path=./model

Run gRPC client

./predict_client.py --host 127.0.0.1 --port 9000 --model_name deep_cnn --model_version 1

Notice that cloudml is not released now.

cloudml models predict -n deep_cnn -s 127.0.0.1:9000 -f ./data.json