paper, show and tell, Google, CVPR 2015
This is a demo for image caption.
I don't create the evaluation code.
If you want to use it, first using download.sh to get the data needed. It will automatically create a directory and download the coco data.
Then using the build_vocab.py to build the vocabulary we need.
After that, use the resize.py to resize all the pictures into the same size. Meanwhile, I also recommend you to use it to resize the test picture.
What's more, use the train.py to train the model. Please adjust the configure by adding the flag or just revise the code for both num_epochs and batch_size。
In the end, you can use sample.py to create the demos for image caption. You may create a directory like 'test' and then add the demo picture into it. Please don't forget to use the resize.py to adjust the size of the pictures.
I use this code in Google Drive colab with free GPU and even TPU in 12 hours continully.