Open bjuncek opened 5 months ago
The "train" and "eval" data segments on GitHub correspond to the "meta-personalization" and "test-time personalization" data discussed in the paper. For query-time retrieval, you can refer to the CSV file we've prepared and shared on GitHub here: https://github.com/danielchyeh/this-is-my/blob/main/this-is-my-dataset/this-is-my_eval-captions.csv
From the README/file structure, it is hard to tell which data corresponds to "meta-personalisation" dataset, which one to "test-time personalisation" (low-shot train, I guess), and which one to query dataset.
It would be helpful if there was a description as to what matches with what specifically.