Repository navigation
Expand file tree
/
Copy pathwaterbirds.py
More file actions
144 lines (112 loc) · 5.23 KB
/
Copy pathwaterbirds.py
File metadata and controls
144 lines (112 loc) · 5.23 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
#!/usr/bin/env python3
# https://nlp.stanford.edu/data/dro/waterbird_complete95_forest2water2.tar.gz
from pathlib import Path
import numpy as np
import torch
import os
from torchvision import datasets
from torchvision import transforms
from PIL import Image
import matplotlib
matplotlib.use("Agg")
from matplotlib import pyplot as plt
from torch.utils.data import Dataset, DataLoader
from typing import List, Callable, Tuple, Generator, Union
from collections import OrderedDict
from torch.utils.data import ConcatDataset
import pandas as pd
import requests
from tqdm import tqdm
import tarfile
#os.chdir('./Aracnonet/')
data_transform = transforms.Compose([
transforms.Resize((224, 224), interpolation=transforms.InterpolationMode.BICUBIC),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)),
])
class Waterbirds(Dataset):
DOWNLOAD_URL = "https://nlp.stanford.edu/data/dro/waterbird_complete95_forest2water2.tar.gz"
DATASET_NAME = "waterbirds"
def __init__(self, env: str, root: str = "./data", transform = data_transform, metadata_filename: str = "metadata.csv", return_index: bool = True):
self.root: str = os.path.join(root, Waterbirds.DATASET_NAME)
self.env: str = env
self.metadata_filename: str = metadata_filename
self.return_index: bool = return_index
self.num_classes = 2
self.env_to_split = {
"train": 0,
"val": 1,
"test": 2
}
if "waterbirds" not in os.listdir("./data"):
self.__download_dataset()
else: self.root = "./data/waterbirds/waterbird_complete95_forest2water2"
self.transform = transform
self.metadata_path = os.path.join(self.root, self.metadata_filename)
metadata_csv = pd.read_csv(self.metadata_path)
metadata_csv = metadata_csv.query(f"split == {self.env_to_split[self.env]}")
self.samples = {}
self.files_count = 0
for i, (_, sample_info) in enumerate(metadata_csv.iterrows()):
self.samples[i] = {
"img_id": sample_info["img_id"],
"image_path": os.path.join(self.root, sample_info["img_filename"]),
"class_label": int(sample_info["y"]),
"bias_label": int(sample_info["place"]),
"all_attrs": list((str(e) for e in sample_info))
}
self.files_count += 1
def __download_dataset(self) -> None:
os.makedirs(self.root, exist_ok=True)
output_path = os.path.join(self.root, "waterbirds.tar.gz")
print(f"=> Downloading {Waterbirds.DATASET_NAME} for {Waterbirds.DOWNLOAD_URL}")
try:
response = requests.get(Waterbirds.DOWNLOAD_URL, stream=True)
response.raise_for_status()
with open(output_path, mode="wb") as write_stream, tqdm(
desc=output_path,
total=int(response.headers["content-length"], 0),
unit="B",
unit_scale=True,
unit_divisor=1024
) as pbar:
for chunk in response.iter_content(chunk_size=8192):
write_stream.write(chunk)
pbar.update(len(chunk))
except:
raise RuntimeError("Unable to complete dataset download, check for your internet connection or try changing download link.")
print(f"=> Extracting waterbird_complete95_forest2water2.tar.gz to directory {self.root}")
try:
with tarfile.open(output_path, mode="r:gz") as unballer:
unballer.extractall(self.root)
except:
raise RuntimeError(f"Unable to extract {output_path}, an error occured.")
self.root = os.path.join(self.root, "waterbird_complete95_forest2water2")
os.remove(output_path)
def __len__(self) -> int:
return self.files_count
def __getitem__(self, index: Union[int, slice, list]) -> Tuple[torch.Tensor]:
if isinstance(index, slice):
return [self.__getitem__(i) for i in range(*index.indices(len(self)))]
if isinstance(index, list):
return [self.__getitem__(idx) for idx in index]
image = self.transform(Image.open(self.samples[index]["image_path"]))
class_label = self.samples[index]["class_label"]
bias_label = self.samples[index]["bias_label"]
img_id = self.samples[index]["img_id"]
return image, (class_label, bias_label), index
def perclass_populations(self, return_labels: bool = False) -> Union[Tuple[float, float], Tuple[Tuple[float, float], torch.Tensor]]:
labels: torch.Tensor = torch.zeros(len(self))
for i in range(len(self)):
labels[i] = self[i][1][0]
_, pop_counts = labels.unique(return_counts=True)
if return_labels:
return pop_counts.long(), labels.long()
return pop_counts
def get_bias_labels(self) -> Generator[None, None, torch.Tensor]:
for i in range(len(self)):
yield self[i][2]
def __repr__(self) -> str:
return f"Waterbirds(env={self.env}, bias_amount=Fixed, num_classes={self.num_classes})"
if __name__ == "__main__":
Waterbirds(env="train")