Hi,
Thanks for sharing your implementation. I notice that the val1/val2 splits are used for training and testing on ImageNet follow the RCNN.
I would ask do you use the images from train set? Because rbgirshick mentioned that images from val1 and 1K images per class from train are both utilized.
Hi,
Thanks for sharing your implementation. I notice that the
val1/val2splits are used for training and testing on ImageNet follow the RCNN.I would ask do you use the images from
trainset? Because rbgirshick mentioned that images fromval1and 1K images per class fromtrainare both utilized.