chenzhi1992 / TensorRT-SSD

Use TensorRT API to implement Caffe-SSD, SSD(channel pruning), Mobilenet-SSD
251 stars 84 forks source link

PreluPlugin constructor from nvinfer1::Weights #8

Open GarrickLin opened 6 years ago

GarrickLin commented 6 years ago

I found that PreluPlugin still need a constructor for nvinfer1::Weights to init prelu parameters

chenzhi1992 commented 6 years ago

You can look at the tensorrt3.0 document.There is an introduction in the document.

nyyznyyz1991 commented 6 years ago

Thanks for your contribution to TensorRT implementation @chenzhi1992 , I found that TensorRT3 already implemented createPReLU INvPlugin * createPReLUPlugin(float negSlope); INvPlugin * createPReLUPlugin(const void * data, size_t length); However it only supports channel_shared mode, so what we need to do is to extend it to cross_channel mode, is it right?

GarrickLin commented 6 years ago

it it not a real PReLU as caffe, it is a leakyrelu with constant neg-slop.