search

kevinlin311tw / Caffe-DeepBinaryCode

Supervised Semantics-preserving Deep Hashing (TPAMI18)
https://arxiv.org/abs/1507.00101v2
Other
205 stars 83 forks source link

About the model used on SUN397 dataset #26

Closed fanwenyuan closed 6 years ago

fanwenyuan commented 6 years ago

Excuse me, when I was trying to train my own model on Sun397 but failed. Did you use a pre-trained model with another dataset?

kevinlin311tw commented 6 years ago

Thanks for your interest in this work. As I can recall, we didn't train the model from scratch. Training from scratch may need additional tricks, and it is not the major scope of this work.

Instead, we trained the model using pre-trained weights from ILSVRC12. This leads to good performance. In addition, we empirically found that training the model using pre-trained weights from Places dataset leads to higher accuracy. I think this is mainly because Places datasets consists of scene images, where should better fit to the image categories in SUN397.