launcher.py

import json
import ntpath
import os

#return from the input file only dict of id & name
def conv_input(input_data):
    id_img = dict()
    for entry in input_data['images']:
        id_img[entry['id']] = entry['file_name']
    return id_img


def search(name, output):
    return [element for element in output if getFilename(element['img_fn']) == name]

#Get file name from the image_fn which include path
def getFilename(path):
    head, tail = ntpath.split(path)
    return tail or ntpath.basename(head)

#Adjust the output of Captionme with the submission output format
def covert_output (input_dic , output_dict):
    results=[]
    for key , value in input_dic.items():
        name = value
        e = search(name, output_dict)
        if (len(e) > 0):
            results.append({'image_id' : key , 'caption' : e[0]['caption']})
            
    # write the results for submission
    with open('pred/results.json', 'w') as file:
         file.write(json.dumps(results))


assert os.path.isfile("input/input.json"), "Input JSON missing!"

with open ('input/input.json') as f2:
    input_data = json.load(f2)

# call the caption generator
os.system('python2 captionme.py')

# read the output of captionme.py
with open ('images_captions.json') as f1:
    captionme_dict = json.load(f1)

# generate the output for submission
adj_input= conv_input(input_data)
covert_output(adj_input, captionme_dict)