Tensorflow Pretrained Dual Path Networks (DPN)

Easy and straightforward tensorflow implementation of Dual-Path-Networks ((https://arxiv.org/abs/1707.01629) with cypw's pretrained weights, converted to tensorflow format.

Details

• All models require 224x224 image size.
• The input images are preprocessed by substracting the RGB mean = [ 124, 117, 104 ], and then multiplying by 0.0167.
• Axis = 3 is considered the channel axis.

How to load weights example

model_type = 'dpn92'
model = dpn_model(input_shape=(224, 224, 3), model_type=model_type)
model.load_weights('{}.h5'.format(model_type))

Pretrained Weights

Model	Weights
DPN-68	https://drive.google.com/file/d/1AYeFy4JZVINqv8MOyg9CwGPT9-f3jFI3/view
DPN-92	https://drive.google.com/file/d/1Ug_rQhxqa9x8vCoaTaLL0geTqbyHEbNk/view
DPN-98	https://drive.google.com/file/d/1Zbrngvbqq6kvop_BCu72Y-BC0DNPtHuB/view
DPN-107	https://drive.google.com/file/d/1_Qkpolqwg9tMqjx6yPD8-p46Hlzneh3L/view
DPN-131	https://drive.google.com/file/d/1kxQyhsLRcIq1aeKt4vUbS1fjfqzg2bpC/view