Skip to content

Latest commit

 

History

History
12 lines (8 loc) · 960 Bytes

README.md

File metadata and controls

12 lines (8 loc) · 960 Bytes

DenseNet-ImageNet pure tensorflow !

Introduction

We convert the caffemodel to .npy weights file (now only DenseNet_161 is available)

For more details about this network, visit this

something about the labels

Note that there two label strategy for ImageNet data. One is the official label, see DenseNet_161/resources/official_order.txt and the other is WNID ascending order, see DenseNet_161/resources/ascending_order.txt. Actually, the latter is much more popular than the former. Most of the weights pretrained on ImageNet use the WNID ascending order label strategy. I mention this problem just to show the differnece between two label strategy. (also because I use the the official label and I am confused at first. I load weights and predict a label for an image but all wrong)

download

DenseNet_161 weights file BaiduDisk