the result of a tensorrt tensor is wrong when the conv pad is different on two axis.here is the layer
layer {
name: "dd"
type: "Convolution"
bottom: "d2"
top: "d3"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
convolution_param {
num_output: 512
stride: 1
weight_filler {
type: "msra"
}
pad_h: 1
pad_w: 0
kernel_h: 1
kernel_w: 1
}
}
this previous layer of this layer is concat.the result of tensor is right in x86 devices,but is wrong in px2.Does nvidia support the conv like this?