I run the duke dataset on 1 GTX2080ti, then I set the parameters as below:
batch_size=64
num_instances=4
iters=200
pooling-type=gem
momentum=0.1
use-hard
lr=0.00035
the other parameters are same as before, finally I got mAP 70.3%, (< mAP 72.8% in paper).
Is it any chance to reach 72.8% when I only use 1 GTX 2080ti?
I run the duke dataset on 1 GTX2080ti, then I set the parameters as below: batch_size=64 num_instances=4 iters=200 pooling-type=gem momentum=0.1 use-hard lr=0.00035 the other parameters are same as before, finally I got mAP 70.3%, (< mAP 72.8% in paper). Is it any chance to reach 72.8% when I only use 1 GTX 2080ti?