search

imatge-upc / detection-2016-nipsws

Hierarchical Object Detection with Deep Reinforcement Learning
http://imatge-upc.github.io/detection-2016-nipsws/
MIT License
423 stars 129 forks source link

Yes, pretrained CNN is used here,using pretrained cnn (vgg16 here) has 2 benefits:First, #27

Open zhongkedazhouman opened 5 years ago

zhongkedazhouman commented 5 years ago

Yes, pretrained CNN is used here,using pretrained cnn (vgg16 here) has 2 benefits:First, learning the Q function is faster because we need to up-date the parameters of the Q-Network only, while using the deep CNN just as a feed-forward feature extractor. Second, the hierarchy of features is trained with a larger dataset,leveraging generic discriminative features in this method.

Originally posted by @mani5h in https://github.com/imatge-upc/detection-2016-nipsws/issues/19#issuecomment-377959805

zhongkedazhouman commented 5 years ago

Thank you for your code. I have two questions.

1. Is the vgg16_weights.h5, which downloaded from the link provided by you, pretrained  on ImageNet?

2. It seems that the model_vgg is trained at every epoch, but do not be saved. At test time, the weights of model_vgg is still the Initial weights(i.e. vgg16_weights.h5). Why not save it and use it at test time?