balancap / SSD-Tensorflow

Single Shot MultiBox Detector in TensorFlow
4.11k stars 1.89k forks source link

Some problems about pool5/关于block5中pool的问题 #276

Open MioChiu opened 6 years ago

MioChiu commented 6 years ago

net = slim.max_pool2d(net, [3, 3], stride=1, scope='pool5') The stride of pool5 is 1 and the size of it is [3,3],the default padding of slim.max_pool2d is 'VALID',why dont author set up padding?Without padding the size of output is [17,17] ,not [19,19]. 第五层pool层中步长为1,size为33 ,slim.max_pool2d默认的是padding='VALID',这样算出来是1717, 但是我们在第七层的feature map要求是19*19啊,这里是否应该设置padding= 'SAME'

bayonettablack commented 6 years ago

In ssd_vgg_300.py, the function ''ssd_arg_scope(...)'' has been set padding = ‘SAME’.