text embeddings preparation

[SreenijaK]

I want to train with my own data and to prepare my dataset and for that I'll have to convert text files to .t7 format for text embeddings. How do i achieve the same?

[shahhaard47]

I've submitted a pull request with requirements.txt file with the correct versions to be used to use this code and work with .t7 files. You can install those versions, like I did, to get it running quickly.

However, this comment describes how to update this code to use latest pytorch version. Other comments on this issue also state that load_lua used on older code is no longer supported and that you might be able to use torchfile.

Furthermore, to get it working with your own dataset, you may need to refer to the original repository and train your own text encoder for the captions.