Reproducing the performance of Referring Image Segmentation task

[golddohyun]

Hi, first of all, thank you for the great and interesting work!
I am struggling to reproduce the G-ref (RefCOCOg) reported performance, and I need your suggestions and detailed info.
I used the exact config provided in the repo and got a val mIoU of 63.02 and oIoU of 60.06 in stead of 66.42 and 63.12 in your paper.
Could you provide more details such as the exact number of iterations, number of GPUs you used for training, global batch size ( i guess in the config the local? batch size is 4) and the version of the libraries, and the exact training time.

Thank you!