/ImageQA_LAB

An LSTM neural network used to proceed image QA, implemented with Tensorflow.

Primary LanguagePython

Speech Lab Project - ImageQA using TensorFlow

NTU CSIE B03902012 Vincent Huang B03902090 Nick Shao

Tensorboard usage

SSH tunneling

Tensorboard is set to run at localhost:6006 by default. To launch from remote server, we have to do ssh tunneling. Here's an example:

ssh -L localhost:16006:localhost:6006 b03902012@server.ip

16006 is any port you like.

Launch tensorboard

logdir

Tensorboard will go to the logdir specified in the source code to find the graph we want to visualize. In train_gensim.py, the default logdir is set to be ./tensorboard/. It can be specified by:

python train_gensim.py --logdir=path/to/logdir/

Launch

tensorboard --logdir=path/to/logdir/

Make sure that the path/to/logdir/ is consistent to the one specified in the execution of the training process. That is to say, Even if you didn't specify the logdir explicitly, you have to use tensorboard --logdir=./tensorboard/ to run properly. Some useful arguments:

  • --debug: print debug messages

Then, we can navigate the local web browser to localhost:6006 to view the TensorBoard.

For more information, tips and debugging, please refer to the TensorBoard README.