Dear author
according to the source code and train log, the batch size is 4x4, so I use the batch size to train the coco and found that there are 7393(118287/16) iters in one epoch, but there are 7329 iters in train log, so the train set is preprocessd by some operations?
Dear author according to the source code and train log, the batch size is 4x4, so I use the batch size to train the coco and found that there are 7393(118287/16) iters in one epoch, but there are 7329 iters in train log, so the train set is preprocessd by some operations?