About MSCOCO dataset

[musicman217]

Hello author, thank you for your excellent work.

Could you please provide several annotation files of MSCOCO? Based on the dataset directory structure you provided, I extracted train_ids and test_ids using the captions_train2014.json and captions_val2014.json files. However, the actual training performence is very poor. I'm wondering if it's because the subset I chose is incorrect. Looking forward to your reply, and thank you again!

[leolee99]

Thank you for your interests to our work, the annotation file could be found here. If the performance is poor, whether if it is caused by torch version? There are great difference between torch 1.X and torch 2.X. If you used torch 2.X, I suggest you try it on torch 1.X again, or at least torch 2.0.