Run the following steps to implement image captioning using a pre-trained im2txt model on TensorFlow:
- The downloaded pre-trained weights checkpoint file (model.ckpt-2000000) contains some issues and you will most likely fail to load it. In that case, fix the issue in the checkpoint by modifying it and creating a new checkpoint file (model2.ckpt-2000000) inside the models folder using code from https://github.com/tensorflow/models/issues/466 or https://github.com/PacktPublishing/Python-Image-Processing-Cookbook/blob/master/Chapter%2009/Chapter%2009.ipynb.
- Build the inference graph from the modified checkpoint file and then load the vocabulary for LSTM (a particular type of RNN that can learn long-term dependencies):
checkpoint_path ...