search

chenzhi1992 / TensorRT-SSD

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

Last layer of network #1

Closed GraphicsHunter closed 6 years ago

GraphicsHunter commented 6 years ago

hello. i saw that u used createSSDDetectionOutputLayer in pluginimplement.cpp to declare the last layer of SSD, but in the deploy prototxt you commented out that layer. Can you explain why? Are you able to release an sample of how to run this code?

chenzhi1992 commented 6 years ago

@tianfanzhu, hi, I wrote the createSSDDetectionOutputLayer function in pluginimplement.cpp , but I didn't use it. Because there's going to be a mistake，if you use this function. I reimplemented this function by using the outputs of mbox_loc, mbox_prior, mbox_conf layer . I'll publish the code later. Please pay attention to my github.

GraphicsHunter commented 6 years ago

Thank you, I will look forward to it. Are you, by the way, able to approximate when this is fully released? Also, how does SSD(channel pruning) work? Is it something that is to be implemented, or is this a procedure that is done during training?

hectorgui commented 5 years ago

@tianfanzhu, hi, I wrote the createSSDDetectionOutputLayer function in pluginimplement.cpp , but I didn't use it. Because there's going to be a mistake，if you use this function. I reimplemented this function by using the outputs of mbox_loc, mbox_prior, mbox_conf layer . I'll publish the code later. Please pay attention to my github.

Hi chenzhi, have you ever upload your new code?