search

zhaoweicai / cascade-rcnn

Caffe implementation of multiple popular object detection frameworks
1.04k stars 293 forks source link

I wonder why the bbox_pre layer's num_output is always 8 #50

Closed Peng-wei-Yu closed 6 years ago

Peng-wei-Yu commented 6 years ago

In the train.prototxt, show as follow: layer { name: "bbox_pred" type: "InnerProduct" bottom: "data" top: "bbox_pred" param { lr_mult: 1 } param { lr_mult: 2 } inner_product_param { num_output: 8 weight_filler { type: "gaussian" std: 0.001 } bias_filler { type: "constant" value: 0 } } } I don't understand why the num_output is 8

zhaoweicai commented 6 years ago

It is class-agnostic regression, the output dim = 2*4.

Peng-wei-Yu commented 6 years ago

@zhaoweicai I search the C++ code and see what does is mean, but why class-agnostic regression. Will is be much more robust?