Some question for reproduce DSSD

[niluanwudidadi]

Hello, chengyangfu. I'm a graduate student. I want to reproduce your achievement. In this process, I have met some problem, for DSSD300, I wonder all the prediction module consist of conv(256), conv(256), conv(1024)? And the all of the deconvolution is 512 dimension? Original SSD's conv4_3(512), fc7(1024), conv8_2(512), conv9_2(256), conv10_2(256), conv11_2(256) , is all of this layer's deconvolution's dimension is 512?
Thank for your work, look forward your reply.