When we load a base network, the weight of the network is meaningless, we only use its structure?

weiliu89 / caffe

Caffe: a fast open framework for deep learning.

http://caffe.berkeleyvision.org/

Other

4.77k stars 1.67k forks source link

When we load a base network, the weight of the network is meaningless, we only use its structure? #497

Open song6299 opened 7 years ago

song6299 commented 7 years ago

When we load the base network such as VGG, is the weight of the pretrained network useful? I think it is meaningless to pretrain the network with our own database. For example, if I want to detect airplanes, I use many airplane slice images to pretrain the network, and then use the trained model as base network of SSD. The method cannot increase detection_eval. Because I didn't find it load the weight of the network in solver.prototxt. Am I right?

song6299 commented 7 years ago

@weiliu89 can you help me, please?