Google Colaboratory 上で Chainer 使って、学習済みの Caffe モデルを使う練習

oxford102.caffemodel のダウンロード

花の画像から、花の名前を予測する Caffe モデルです

# URL によるリソースへのアクセスを提供するライブラリをインポートする。
import urllib.request 
# ウェブ上のリソースを指定する
url = 'https://s3.amazonaws.com/jgoode/oxford102.caffemodel'
# 指定したURLからリソースをダウンロードし、名前をつける。
urllib.request.urlretrieve(url, 'oxford102.caffemodel')

('oxford102.caffemodel', <http.client.HTTPMessage at 0x7ff854b81b00>)

Chainer による Caffe モデルの利用

from chainer.links import caffe
func = caffe.CaffeFunction('oxford102.caffemodel')

/usr/local/lib/python3.6/dist-packages/chainer/links/caffe/caffe_function.py:174: UserWarning: Skip the layer "data", since CaffeFunction does not support it
  'support it' % layer.name)

画像処理してから予測まで

import urllib.request 
from PIL import Image
import numpy as np
import chainer
import chainer.functions as F

filename = "whatisyourname.jpg"
def guess_flower_name(url, filename=filename):
    labels = ['pink primrose', 'hard-leaved pocket orchid', 'canterbury bells', 'sweet pea', 
              'english marigold', 'tiger lily', 'moon orchid', 'bird of paradise', 'monkshood', 
              'globe thistle', 'snapdragon', "colt's foot", 'king protea', 'spear thistle', 
              'yellow iris', 'globe-flower', 'purple coneflower', 'peruvian lily', 'balloon flower', 
              'giant white arum lily', 'fire lily', 'pincushion flower', 'fritillary', 'red ginger', 
              'grape hyacinth', 'corn poppy', 'prince of wales feathers', 'stemless gentian', 
              'artichoke', 'sweet william', 'carnation', 'garden phlox', 'love in the mist', 
              'mexican aster', 'alpine sea holly', 'ruby-lipped cattleya', 'cape flower', 
              'great masterwort', 'siam tulip', 'lenten rose', 'barbeton daisy', 'daffodil', 
              'sword lily', 'poinsettia', 'bolero deep blue', 'wallflower', 'marigold', 
              'buttercup', 'oxeye daisy', 'common dandelion', 'petunia', 'wild pansy', 
              'primula', 'sunflower', 'pelargonium', 'bishop of llandaff', 'gaura', 'geranium', 
              'orange dahlia', 'pink-yellow dahlia?', 'cautleya spicata', 'japanese anemone', 
              'black-eyed susan', 'silverbush', 'californian poppy', 'osteospermum', 
              'spring crocus', 'bearded iris', 'windflower', 'tree poppy', 'gazania', 'azalea', 
              'water lily', 'rose', 'thorn apple', 'morning glory', 'passion flower', 'lotus', 
              'toad lily', 'anthurium', 'frangipani', 'clematis', 'hibiscus', 'columbine', 
              'desert-rose', 'tree mallow', 'magnolia', 'cyclamen ', 'watercress', 'canna lily', 
              'hippeastrum ', 'bee balm', 'ball moss', 'foxglove', 'bougainvillea', 'camellia', 
              'mallow', 'mexican petunia', 'bromelia', 'blanket flower', 'trumpet creeper', 
              'blackberry lily']
    urllib.request.urlretrieve(url, filename)
    image = Image.open(filename).convert('RGB')
    w, h = image.size
    fixed_w, fixed_h = 224, 224
    if w > h:
        shape = (int(fixed_w * w / h), fixed_h)
    else:
        shape = (fixed_w, int(fixed_h * h / w))

    left = (shape[0] - fixed_w) / 2
    right = left + fixed_w
    top = (shape[1] - fixed_h) / 2
    bottom = top + fixed_h

    image = image.resize(shape)
    image = image.crop((left, top, right, bottom)) 

    x_data = np.asarray(image).astype(np.float32)
    x_data = x_data.transpose(2, 0, 1)
    x_data = x_data[::-1, :, :]

    mean_image = np.zeros(3*fixed_w*fixed_h).reshape(3, fixed_w, fixed_h).astype(np.float32)
    mean_image[0] = 103
    mean_image[1] = 117
    mean_image[2] = 123

    x_data -= mean_image
    x_data = np.array([x_data])

    x = chainer.Variable(x_data)
    func = caffe.CaffeFunction('oxford102.caffemodel')
    y, = func(inputs={'data': x}, outputs=['fc8_oxford_102'])

    prob = F.softmax(y)

    for i in range(5):
        idx = np.argsort(prob.data[0])[::-1][i]
        print(i + 1, "\t", labels[idx], "\t", prob.data[0][idx])

    return image

guess_flower_name('https://d1f5hsy4d47upe.cloudfront.net/93/936a7824f1041bfef2cbe7d96c1fc7cc_t.jpeg')

/usr/local/lib/python3.6/dist-packages/chainer/links/caffe/caffe_function.py:174: UserWarning: Skip the layer "data", since CaffeFunction does not support it
  'support it' % layer.name)


1 	 carnation 	 0.6862433
2 	 wallflower 	 0.075164974
3 	 corn poppy 	 0.057577945
4 	 hibiscus 	 0.037174866
5 	 great masterwort 	 0.033413775

url = "https://c8.alamy.com/comp/HMWBKR/crimson-cattleya-or-ruby-lipped-cattleya-cattleya-labiata-candida-HMWBKR.jpg"
guess_flower_name(url)

/usr/local/lib/python3.6/dist-packages/chainer/links/caffe/caffe_function.py:174: UserWarning: Skip the layer "data", since CaffeFunction does not support it
  'support it' % layer.name)


1 	 carnation 	 0.56166863
2 	 sword lily 	 0.4131709
3 	 camellia 	 0.0039400724
4 	 bearded iris 	 0.003937593
5 	 ruby-lipped cattleya 	 0.003253354