Dataset and the running sequence of the code

[ExitPath]

Hello, I have read your paper, how interesting it is! Yet I have some questions when reproducing this work. 1. The download link of the dataset and how should it be preprocessed? 2. What is the running sequence of the code? Could you please introduce it more clearly in the Readme. thank you very much！

[ezeli]

Sorry, I still have some expansion work in progress, so I can't provide it in full.
However, the dataset and a lot of preprocessing work can refer to my other repo and preprocess.py.
Running sequence: train_cpt.py, train_senti.py and train_sent_senti_cls_rnn.py -> train_xe.py -> train_rl.py.
When I finish all the work, I will provide all the data.

[terence1023]

Sorry, I still have some expansion work in progress, so I can't provide it in full. However, the dataset and a lot of preprocessing work can refer to my other repo and preprocess.py. Running sequence: train_cpt.py, train_senti.py and train_sent_senti_cls_rnn.py -> train_xe.py -> train_rl.py. When I finish all the work, I will provide all the data.

Hello, I am also interested in this paper! Excuse me, can all the data be released now? Thank you very much for your help!

[ezeli]

Sorry, our extended paper has not been accepted yet. The open source datasets mentioned in the paper can be downloaded by yourself. The sentiment dataset we label for coco is only used to verify the performance of the sentiment classifier and can be omitted. All other code is already in this repository.

[ezeli]

Note: The image sentiment datasets are only used for train_senti.py, not for train_rl.py. The rl_senti_times parameter in train_rl.py is set to 0.
The preprocessing code includes preprocess.py and detect_concepts.py. detect_concepts.py is used to detect the concept word of the image after training the concept detector.
So the dataset we don't currently provide (just to verify the performance of the sentiment detector on the coco dataset) doesn't actually affect the use of this repository.