This repo is for Paper Contrastive Learning Based Visual Representation Enhancement for Multimodal Machine Translation
- First of all, the image data needs to be downloaded by person.
- With a request to https://forms.illinois.edu/sec/229675, you will get the Flickr30k image dataset, which contains images for train, valid and test_2016_flickr.
- Test_2017_flickr is in https://drive.google.com/drive/folders/1kfgmYFL5kup51ET7WQNxYmKCvwz_Hjkt
- MSCOCO is in http://images.cocodataset.org/#download
- Test_2018_flickr is in https://drive.google.com/drive/folders/1kfgmYFL5kup51ET7WQNxYmKCvwz_Hjkt
The way of processing image data is written in 'process_imgdata.sh' Run the shell file, all image data will be the same as we used in our work
- 'run.sh' is the code for running our code. It contains how to preprocess text data, train, inference and evaluation