Upload process_dataset.py

scripts/process_dataset.py

@@ -1,70 +1,35 @@
-"""This script processes the dataset provided by the PathVQA authors. This script drops the duplicate
-image-question-answer triplets, converts the images to bytes, and saves the dataset to a parquet file.
+"""This script de-duplicates the data provided by the PathVQA authors,
+creates an "imagefolder" dataset and pushes it to the hub.
 """
 
-import io
+import os
+import shutil
 import pickle
+import datasets
 import pandas as pd
-from PIL import Image
-from tqdm import tqdm
 
-# loop across the splits
 for split in ["train", "val", "test"]:
     
+    os.makedirs(f"data/{split}/", exist_ok=True)
+    
     # load the image-question-answer triplets
     data = pd.DataFrame(pickle.load(open(f"pvqa/qas/{split}/{split}_qa.pkl", "rb")))
-    print(f"Total number of triplets in {split} set: {format(data.shape[0], ',.0f')}")
     
     # drop the duplicate image-question-answer triplets
     data = data.drop_duplicates(ignore_index=True)
-    print(f"Unique number of triplets in {split} set: {format(data.shape[0], ',.0f')}")
     
-    # load the images as bytes
-    print(f"Loading {split} set images...")
-    images = {}
-    for image in tqdm(data["image"].unique()):
-        img = Image.open(f"pvqa/images/{split}/{image}.jpg")
-        byt = io.BytesIO()
-        img.save(byt, format="jpeg")
-        byt = byt.getvalue()
-        images[image] = {"path": None, "bytes": byt}
-    print(f"Unique number of images in {split} set: {format(len(images), ',.0f')}")
+    # convert the image names to file names
+    data = data.rename(columns={"image": "file_name"})
+    data["file_name"] += ".jpg"
     
-    # save the data to a parquet file
-    print(f"Writing data to data/{split}.parquet...")
-    dataset = []
-    for _, row in data.iterrows():
-        dataset.append({
-            "image": images[row["image"]],
-            "question": row["question"],
-            "answer": row["answer"]
-        })
-    pd.DataFrame(dataset).to_parquet(f"data/{split}.parquet")
-    print("Done")
-    print("---------------------------------")
+    # copy the images referenced by the question-answer pairs
+    for image in data["file_name"].unique():
+        shutil.copyfile(src=f"pvqa/images/{split}/{image}", dst=f"data/{split}/{image}")
     
-'''
-Total number of triplets in train set: 19,755
-Unique number of triplets in train set: 19,654
-Loading train set images...
-100%|██████████| 2599/2599 [00:46<00:00, 56.27it/s]
-Unique number of images in train set: 2,599
-Writing data to data/train.parquet...
-Done
----------------------------------
-Total number of triplets in val set: 6,279
-Unique number of triplets in val set: 6,259
-Loading val set images...
-100%|██████████| 832/832 [00:13<00:00, 59.49it/s]
-Unique number of images in val set: 832
-Writing data to data/val.parquet...
-Done
----------------------------------
-Total number of triplets in test set: 6,761
-Unique number of triplets in test set: 6,719
-Loading test set images...
-100%|██████████| 858/858 [00:15<00:00, 53.93it/s]
-Unique number of images in test set: 858
-Writing data to data/test.parquet...
-Done
-'''
+    # save the metadata
+    data.to_csv(f"data/{split}/metadata.csv", index=False)
+
+# push the dataset to the hub
+dataset = datasets.load_dataset("imagefolder", data_dir="data/")
+dataset.push_to_hub("flaviagiammarino/path-vqa")
+