organize mini-ImageNet files for MAML training

prepare_miniimagenet.py

#!/usr/bin/env python
import csv
from PIL import Image
import pickle
import os

img_size = 84

test_csv_file = "../../Code/few-shot-ssl-public/fewshot/data/mini_imagenet_split/Ravi/test.csv"
train_csv_file = "../../Code/few-shot-ssl-public/fewshot/data/mini_imagenet_split/Ravi/train.csv"
val_csv_file = "../../Code/few-shot-ssl-public/fewshot/data/mini_imagenet_split/Ravi/val.csv"

test_pkl = "mini-imagenet-cache-test.pkl"
train_pkl = "mini-imagenet-cache-train.pkl"
val_pkl = "mini-imagenet-cache-val.pkl"

def pkl_to_raw(csv_file, pkl_file, dirname):
  if not os.path.isdir(dirname):
    os.mkdir(dirname)
    print("made: ", dirname)
  p = pickle.load(open(pkl_file, 'rb'))
  imgs = p["image_data"]
  csv_r = csv.reader(open(csv_file))
  i = -1
  last_label = ""
  for (image_filename, class_name) in csv_r:
    # skip headers
    if i == -1:
      i += 1
      continue

    im = Image.fromarray(imgs[i])
    # resize as in maml source code
    im = im.resize((img_size, img_size), resample=Image.LANCZOS)
    
    # make a new subdir for every label
    if class_name != last_label:
      os.mkdir(dirname + "/" + class_name)
      last_label = class_name
    
    # save image file into correct class folder
    new_filename = dirname + "/" + class_name + "/" +  image_filename
    im.save(new_filename)
    i += 1
    if i % 500 == 0:
      print(i)

pkl_to_raw(train_csv_file, train_pkl, "train")
pkl_to_raw(val_csv_file, val_pkl, "val")
pkl_to_raw(test_csv_file, test_pkl, "test")