Add ids and image ids columns

id_finder.py

from tqdm import tqdm
import string
import json
import collections
import numpy as np
import os

DATA_PATH = "data"

################################################################################################
# IMPORT NUMPY FILES
################################################################################################

np_train_data = np.load(os.path.join(DATA_PATH,'train_data.npy'))
np_val_data = np.load(os.path.join(DATA_PATH,'val_data.npy'))
np_test_data = np.load(os.path.join(DATA_PATH,'test_data.npy'))

train_data = collections.OrderedDict()
for i in range(len(np_train_data.item())):
    cap =  np_train_data.item()['caps']
    img =  np_train_data.item()['ims']
    train_data['caps'] = cap
    train_data['ims'] = img
    
val_data = collections.OrderedDict()
for i in range(len(np_val_data.item())):
    cap =  np_val_data.item()['caps']
    img =  np_val_data.item()['ims']
    val_data['caps'] = cap
    val_data['ims'] = img
    
test_data = collections.OrderedDict()
for i in range(len(np_test_data.item())):
    cap =  np_test_data.item()['caps']
    img =  np_test_data.item()['ims']
    test_data['caps'] = cap
    test_data['ims'] = img


################################################################################################
# IMPORT FILES JSON FILES
################################################################################################

with open(os.path.join(DATA_PATH,'instances_val2014.json')) as json_file:
    coco_instances_val = json.load(json_file)
    
with open(os.path.join(DATA_PATH,'captions_val2014.json')) as json_file:
    coco_caption_val = json.load(json_file)


################################################################################################
# HELPER FUNCTIONS
################################################################################################
    
# Group captions for the same image together
def group_captions(data):
  return np.array(data['caps']).reshape(-1, 5)

# Remove punctuation and make it lowercase
def process_annotations(annotations):
  result = []
  
  for i in annotations:
    translation = ''.join(c for c in i["caption"] if c not in string.punctuation)
    i["processed"] = translation.lower()
    result.append(i)
  
  return result

# Get matches using the captions
def get_matches(captions, lookup):
  result = []
  
  for i in tqdm(range(len(captions))):
    
    found = False
    for caption in captions[i]:
      match = get_match(query=caption, lookup=lookup)
    
      # If it happens that we couldn't find the first caption in the lookup,
      # which happens rarely, we continue to the next one.
      if match:
        found = True
        result.append(match)
        break
      
    if not found:
      result.append(None)
      
  return result

# Get item that matches query
def get_match(query, lookup):
  q = str(query, "utf8")
  
  for i in lookup:
    if q in i["processed"]:
      return i
  return None


################################################################################################
# MAIN FUNCTIONS
################################################################################################

# Use the captions to find the ids
def get_ids(coco_caption, data):
  caps = group_captions(data)
  lookup = coco_caption["annotations"]
  lookup = process_annotations(lookup)
  matches = get_matches(caps, lookup)
  ids = [i["id"] for i in matches]
  image_ids = [i["image_id"] for i in matches]
  return ids, image_ids, caps

# Adding the ids and image ids as columns to data, and group the captions
def add_ids(coco_caption, data):
  ids, image_ids, caps = get_ids(coco_caption, data)
  data["caps"] = caps
  data["ids"] = ids
  data["image_ids"] = image_ids
  return data

################################################################################################
# ADD COLUMNS IDS AND IMAGE IDS, AND GROUP CAPTIONS
################################################################################################

val_data = add_ids(coco_caption_val, val_data)
train_data = add_ids(coco_caption_val, train_data)
test_data = add_ids(coco_caption_val, test_data)
np.save("val_data_with_ids", val_data)
np.save("train_data_with_ids", train_data)
np.save("test_data_with_ids", test_data)

To retrieve the new files:

np_train_data_2 = np.load(os.path.join('train_data_with_ids.npy')).item()
np_val_data_2 = np.load(os.path.join('val_data_with_ids.npy')).item()
np_test_data_2 = np.load(os.path.join('test_data_with_ids.npy')).item()

Example:

np_train_data_2["ids"][0]  # --> 681330