Hi,
i try to run deeplabv3+ (mobilenet encoder) with tensorrt. But there are custom layers like bilinearUpsampling or Relu6 who are not supoorted and i cant run the uff parser. I cant find a good example how to implement custom layers in TensorRT like bilinearUpsampling or Relu6. Is there a support or how can i implement these custom layers? I read about transform to deconv layers instead - but this will enlarge my network?
.pb File Relu6 Layer:
node {
name: "Conv_Relu6/Relu6"
op: "Relu6"
input: "Conv_BN/cond/Merge"
attr {
key: "T"
value {
type: DT_FLOAT
}
}
}
.pb File BilinearUpsampling Layer:
node {
name: "bilinear_upsampling_4/resize_images/ResizeBilinear"
op: "ResizeBilinear"
input: "activation_8/Relu"
input: "bilinear_upsampling_4/resize_images/size"
attr {
key: "T"
value {
type: DT_FLOAT
}
}
attr {
key: "align_corners"
value {
b: false
}
}
}
Thanks for your help in advance!