bentrevett commited on
Commit
4b3cb3f
1 Parent(s): 40e114d

Create script.py

Browse files
Files changed (1) hide show
  1. script.py +67 -0
script.py ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ This is the script used to create the dataset from the downloaded/extracted dataset @ https://www.vision.caltech.edu/datasets/cub_200_2011/
3
+ """
4
+
5
+ import datasets
6
+ from pathlib import Path
7
+ import shutil
8
+ import json
9
+
10
+ index_to_path = (
11
+ Path("CUB_200_2011/CUB_200_2011/images.txt")
12
+ .read_text()
13
+ .strip()
14
+ .split("\n")
15
+ )
16
+ index_to_path = [Path("CUB_200_2011/CUB_200_2011/images") / Path(x.split(" ")[-1]) for x in index_to_path]
17
+
18
+ index_to_split = (
19
+ Path("CUB_200_2011/CUB_200_2011/train_test_split.txt")
20
+ .read_text()
21
+ .strip()
22
+ .split("\n")
23
+ )
24
+ index_to_split = ["train" if x.split(" ")[-1] == "1" else "test" for x in index_to_split]
25
+
26
+ index_to_bbox = (
27
+ Path("CUB_200_2011/CUB_200_2011/bounding_boxes.txt")
28
+ .read_text()
29
+ .strip()
30
+ .split("\n")
31
+ )
32
+
33
+ def convert_bbox(bbox):
34
+ # From: x0, y0, width, height
35
+ # To: x0, y0, x1, y1
36
+ new_bbox = [bbox[0], bbox[1], bbox[0] + bbox[2], bbox[1] + bbox[3]]
37
+ return new_bbox
38
+
39
+ index_to_bbox = [[float(i) for i in x.split(" ")[1:]] for x in index_to_bbox]
40
+ index_to_bbox = [convert_bbox(bbox) for bbox in index_to_bbox]
41
+
42
+ data_dir = Path("data")
43
+ train_dir = Path("data") / Path("train")
44
+ test_dir = Path("data") / Path("test")
45
+ train_dir.mkdir(parents=True, exist_ok=True)
46
+ test_dir.mkdir(parents=True, exist_ok=True)
47
+
48
+ metadata = []
49
+
50
+ for path, split, bbox in zip(index_to_path, index_to_split, index_to_bbox):
51
+ class_dir, file_name = path.parts[-2:]
52
+ dir = train_dir / class_dir if split == "train" else test_dir / class_dir
53
+ dir.mkdir(parents=True, exist_ok=True)
54
+ class_file_path = Path("/".join([class_dir, file_name]))
55
+ destination_path = train_dir / class_file_path if split == "train" else test_dir / class_file_path
56
+ metadata_file_name = Path("/".join(destination_path.parts[1:]))
57
+ x_min, y_min, x_max, y_max = bbox[0], bbox[1], bbox[0] + bbox[2], bbox[1] + bbox[3]
58
+ metadata.append({"file_name": str(metadata_file_name), "bbox": bbox})
59
+ shutil.copy(path, destination_path)
60
+
61
+ with open("data/metadata.jsonl", "w") as f:
62
+ for md in metadata:
63
+ f.write(f"{json.dumps(md)}\n")
64
+
65
+ dataset = datasets.load_dataset("imagefolder", data_dir="data", drop_labels=False)
66
+
67
+ dataset.push_to_hub("bentrevett/cub-200-2011")