Low eval for REC tasks using resnet50 in refcoco #59
Description
I just wanted to implement the task of REC with refcoco with resnet50. I downloaded the refcoco and coco data sets and placed them in the required format shown in the figure. I trained them for the required 20 rounds in the specified single_task_rec.yaml, but the results were terrible. I desperately want to know what I should do
