Using Google Cloud Service for Large Scale Image Labelling

Installing Google SDK

https://cloud.google.com/sdk/docs/quickstart-windows

1. New a Google Cloud Platform Project
2. Download Google Cloud SDK
   1. After installation run gcloud init and log in to your account there!
   2. Select the GCP Project and the computing zone
   3. Finish the SDK configuration!

Installing Google API for different programs (like Vision we use)

https://cloud.google.com/python/

https://googleapis.github.io/google-cloud-python/latest/vision/index.html

conda install -c conda-forge google-cloud-vision

Following packages may be necassary

conda install -c conda-forge google-api-python-client

conda install -c conda-forge google-auth

conda install -c conda-forge google-auth-oauthlib

Set up Credentials for Google on PC

https://cloud.google.com/docs/authentication/getting-started

1. Follow the steps to get an account and a credential. Download the JSON file it generates
2. export or set the environment variables manually
   1. set GOOGLE_APPLICATION_CREDENTIALS=[path_to_json]

Note: If this is not set, then you can import the google packages like from google.cloud import vision. But you will see errors when creating a client!

https://cloud.google.com/docs/authentication/production#obtaining_and_providing_service_account_credentials_manually

Note: If the environment variable is not working, manually set it in python.

import os
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = r"D:\Github\Google_Vision_Image_Labelling\Image-patch-labelling-4e4cfd1664c2.json"

Use the service through the client!

import os
import io
import json
import pickle
from os.path import join
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = r"D:\Github\Google_Vision_Image_Labelling\Image-patch-labelling-4e4cfd1664c2.json"
# set up the client
from google.cloud import vision
client = vision.ImageAnnotatorClient()
#%%
imgbasedir = r"N:\Stimuli\2019-Free-Viewing"
subfolders = ['2019-08-19-Beto',
]
for subfolder in subfolders:
    imgdir = join(imgbasedir, subfolder)
    # filter image names by their suffix
    imgnames = [fn for fn in os.listdir(imgdir) if \
                any(fn.endswith('.'+sfx) for sfx in ["bmp","BMP","jpg","jpeg","png","JPG","JPEG","tiff","TIFF"]) and \
                not os.path.isdir(join(imgdir, fn))]
    print("%d Images in %s folder" % (len(imgnames), subfolder))

#%%
from time import time
t0 = time()
for subfolder in subfolders:
    imgdir = join(imgbasedir, subfolder)
    # filter image names by their suffix
    imgnames = [fn for fn in os.listdir(imgdir) if \
                any(fn.endswith('.'+sfx) for sfx in ["bmp","BMP","jpg","jpeg","png","JPG","JPEG","tiff","TIFF"]) and \
                not os.path.isdir(join(imgdir, fn))]
    print("%d Images in %s folder" % (len(imgnames), subfolder))

    image_parse_dict = {}
    for imgname in imgnames:
        with io.open(join(imgdir, imgname), 'rb') as image_file:
            content = image_file.read()
        print(imgname)
        image = vision.types.Image(content=content)

        response = client.label_detection(image=image)
        labels = response.label_annotations
        print('Labels:')

        label_data = []
        for label in labels:
            print("%s score: %.4f " % (label.description, label.score))
            label_data.append((label.description, label.score))
        print("\n")
        image_parse_dict[imgname] = label_data

    pickle.dump(image_parse_dict, open(join(imgdir, "label_dict.pkl"), "wb")) # Pkl file saved in binary format
    json.dump(image_parse_dict, open(join(imgdir, 'label_result.json'), 'w')) # json file saved in readable format
    print("%.1fs" % (time() - t0))
    # JSON is easier for matlab to read!!
    # do `jsonData = jsondecode(fileread('..\\label_result.json'));` in matlab

print("%.1fs" % (time() - t0))
#%% Loading code post hoc
image_parse_dict = pickle.load(open(os.path.join(imgdir, "label_dict.pkl"), "rb"))
data_dict = json.load(open('result.json', 'r'))