Awesome

chainer-imagenet-vgg

Load caffemodel and transfer it to chainermodel, then save it & use it to predict label of a image.

Requirements

• Chainer is necessary
• If you create chainermodel from caffemodel by yourself,
  • Caffe is required

Download chainermodel

If you don't want to compile caffe, there is chainermodel of VGG-16 converted from caffemodel provided here. It's trained on ILSVRC-2014 dataset.

To get it, just run

$ bash download_chainermodel.sh

Run demo

$ python predict.py

It may produce the below outputs:

['n02124075' 'Egyptian cat']	probability:0.737295150757
['n02123045' 'tabby, tabby cat']	probability:0.17518492043
['n02123159' 'tiger cat']	probability:0.0533684939146
['n02127052' 'lynx, catamount']	probability:0.005824980326
['n04074963' 'remote control, remote']	probability:0.00200812658295

Create chainermodel from caffemodel

Download files

$ bash download_caffemodel.sh

Load caffemodel and save it as chainermodel

$ python load_model.py