retinaface res50 have alittle question
sky186 opened this issue · comments
@clancylian
mobilenet is OK feature have a little not same
res50 model ,the Deconclution shape error
I should know modify the two layer?,
the crop param only inter.
layer {
bottom: "rf_c2_aggr_relu"
top: "rf_c2_upsampling"
name: "rf_c2_upsampling"
type: "Deconvolution"
convolution_param {
num_output: 64
kernel_size: 4
stride: 2
pad: 1
group: 64
bias_term: false
weight_filler: {
type: "bilinear"
}
}
}
layer {
bottom: "rf_c2_upsampling"
bottom: "rf_c1_red_conv_relu"
top: "crop1"
name: "crop1"
type: "Crop"
crop_param {
axis: 1
offset: 0
offset: 0
offset: 0
}
}