Datasets:
Script to sample/download data with requirements.txt
Browse files- download_swim.py +112 -0
- requirements.txt +4 -0
- sample_swim.py +83 -0
download_swim.py
ADDED
@@ -0,0 +1,112 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
from io import BytesIO
|
3 |
+
from pathlib import Path
|
4 |
+
from huggingface_hub import list_repo_tree, hf_hub_url, HfFileSystem
|
5 |
+
from huggingface_hub.hf_api import RepoFile
|
6 |
+
import fsspec
|
7 |
+
from PIL import Image
|
8 |
+
from tqdm import tqdm
|
9 |
+
|
10 |
+
def enumerate_chunks(repo_id, images_parent):
|
11 |
+
"""
|
12 |
+
Lists all immediate chunk subdirs under the images parent using HfFileSystem.
|
13 |
+
Returns sorted list of subdir names (e.g. ['000', '001', ...]).
|
14 |
+
"""
|
15 |
+
fs = HfFileSystem()
|
16 |
+
repo_path = f"datasets/{repo_id}/{images_parent}"
|
17 |
+
entries = fs.ls(repo_path, detail=True)
|
18 |
+
subdirs = [entry['name'].split('/')[-1] for entry in entries if entry['type'] == 'directory']
|
19 |
+
subdirs.sort()
|
20 |
+
return subdirs
|
21 |
+
|
22 |
+
def sample_dataset(
|
23 |
+
repo_id: str,
|
24 |
+
images_parent: str,
|
25 |
+
labels_parent: str,
|
26 |
+
output_dir: str,
|
27 |
+
# max_files: int = 500,
|
28 |
+
flatten: bool = True,
|
29 |
+
chunks: list = None
|
30 |
+
):
|
31 |
+
total_downloaded = 0
|
32 |
+
all_chunks = chunks
|
33 |
+
if all_chunks is None:
|
34 |
+
all_chunks = enumerate_chunks(repo_id, images_parent)
|
35 |
+
print(f"Found chunks: {all_chunks}")
|
36 |
+
for chunk in all_chunks:
|
37 |
+
image_subdir = f"{images_parent}/{chunk}"
|
38 |
+
label_subdir = f"{labels_parent}/{chunk}"
|
39 |
+
|
40 |
+
# List only in the specified chunk
|
41 |
+
image_files = list_repo_tree(
|
42 |
+
repo_id=repo_id,
|
43 |
+
path_in_repo=image_subdir,
|
44 |
+
repo_type="dataset",
|
45 |
+
recursive=True,
|
46 |
+
)
|
47 |
+
|
48 |
+
for img_file in tqdm(image_files, desc=f"Downloading {chunk}", leave=False):
|
49 |
+
if not isinstance(img_file, RepoFile) or not img_file.path.lower().endswith(".png"):
|
50 |
+
continue
|
51 |
+
|
52 |
+
rel_path = Path(img_file.path).relative_to(image_subdir)
|
53 |
+
label_path = f"{label_subdir}/{rel_path.with_suffix('.txt')}"
|
54 |
+
|
55 |
+
if flatten:
|
56 |
+
parts = img_file.path.split('/')
|
57 |
+
# print(parts)
|
58 |
+
# Remove the chunk dir (second last)
|
59 |
+
flat_path = '/'.join(parts[:-2] + [parts[-1]])
|
60 |
+
# For labels, also strip the chunk and substitute extension
|
61 |
+
flat_label_path = flat_path.replace('.png', '.txt').replace('images', 'labels')
|
62 |
+
local_image_path = Path(output_dir) / flat_path
|
63 |
+
local_label_path = Path(output_dir) / flat_label_path
|
64 |
+
else:
|
65 |
+
local_image_path = Path(output_dir) / img_file.path
|
66 |
+
local_label_path = Path(output_dir) / label_path
|
67 |
+
|
68 |
+
local_image_path.parent.mkdir(parents=True, exist_ok=True)
|
69 |
+
local_label_path.parent.mkdir(parents=True, exist_ok=True)
|
70 |
+
|
71 |
+
image_url = hf_hub_url(repo_id=repo_id, filename=img_file.path, repo_type="dataset")
|
72 |
+
label_url = hf_hub_url(repo_id=repo_id, filename=label_path, repo_type="dataset")
|
73 |
+
try:
|
74 |
+
with fsspec.open(image_url) as f:
|
75 |
+
image = Image.open(BytesIO(f.read()))
|
76 |
+
image.save(local_image_path)
|
77 |
+
with fsspec.open(label_url) as f:
|
78 |
+
txt_content = f.read()
|
79 |
+
with open(local_label_path, "wb") as out_f:
|
80 |
+
out_f.write(txt_content)
|
81 |
+
total_downloaded += 1
|
82 |
+
except Exception as e:
|
83 |
+
print(f"Failed {rel_path}: {e}")
|
84 |
+
|
85 |
+
|
86 |
+
|
87 |
+
|
88 |
+
print(f"Downloaded {total_downloaded} image/txt pairs.")
|
89 |
+
print(f"Saved under: {Path(output_dir).resolve()}")
|
90 |
+
|
91 |
+
def parse_args():
|
92 |
+
parser = argparse.ArgumentParser(description="Stream and sample paired images + txt labels from a Hugging Face folder-structured dataset, optionally across multiple chunks.")
|
93 |
+
parser.add_argument("--repo-id", default="JeffreyJsam/SWiM-SpacecraftWithMasks", help="Hugging Face dataset repo ID.")
|
94 |
+
parser.add_argument("--images-parent", default="Baseline/images/val", help="Parent directory for image chunks.")
|
95 |
+
parser.add_argument("--labels-parent", default="Baseline/labels/val", help="Parent directory for label chunks.")
|
96 |
+
parser.add_argument("--output-dir", default="./SWiM", help="Where to save sampled data.")
|
97 |
+
#parser.add_argument("--count", type=int, default=500, help="How many samples to download in total.")
|
98 |
+
parser.add_argument("--flatten", default=True, type=bool, help="Save all samples in a single folder without subdirectories.")
|
99 |
+
parser.add_argument("--chunks", nargs="*", default=None, help="Specific chunk names to sample (e.g. 000 001). Leave empty to process all.")
|
100 |
+
return parser.parse_args()
|
101 |
+
|
102 |
+
if __name__ == "__main__":
|
103 |
+
args = parse_args()
|
104 |
+
sample_dataset(
|
105 |
+
repo_id=args.repo_id,
|
106 |
+
images_parent=args.images_parent,
|
107 |
+
labels_parent=args.labels_parent,
|
108 |
+
output_dir=args.output_dir,
|
109 |
+
# max_files=args.count,
|
110 |
+
flatten=args.flatten,
|
111 |
+
chunks=args.chunks
|
112 |
+
)
|
requirements.txt
ADDED
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
1 |
+
huggingface-hub>=0.23.0
|
2 |
+
fsspec>=2024.6.0
|
3 |
+
Pillow>=10.3.0
|
4 |
+
tqdm>=4.66.4
|
sample_swim.py
ADDED
@@ -0,0 +1,83 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
from io import BytesIO
|
3 |
+
from pathlib import Path
|
4 |
+
from huggingface_hub import list_repo_tree, hf_hub_url
|
5 |
+
from huggingface_hub.hf_api import RepoFile
|
6 |
+
import fsspec
|
7 |
+
from PIL import Image
|
8 |
+
from tqdm import tqdm
|
9 |
+
|
10 |
+
def sample_dataset(
|
11 |
+
repo_id: str,
|
12 |
+
image_subdir: str,
|
13 |
+
label_subdir: str,
|
14 |
+
output_dir: str,
|
15 |
+
max_files: int = 500,
|
16 |
+
):
|
17 |
+
|
18 |
+
image_files = list_repo_tree(
|
19 |
+
repo_id=repo_id,
|
20 |
+
path_in_repo=image_subdir,
|
21 |
+
repo_type="dataset",
|
22 |
+
recursive=True
|
23 |
+
)
|
24 |
+
|
25 |
+
count = 0
|
26 |
+
for img_file in tqdm(image_files, desc="Downloading samples"):
|
27 |
+
if not isinstance(img_file, RepoFile) or not img_file.path.lower().endswith((".png")):
|
28 |
+
continue
|
29 |
+
|
30 |
+
# Relative path after the image_subdir (e.g., img_0001.png)
|
31 |
+
rel_path = Path(img_file.path).relative_to(image_subdir)
|
32 |
+
label_path = f"{label_subdir}/{rel_path.with_suffix('.txt')}" # Change extension to .txt
|
33 |
+
|
34 |
+
image_url = hf_hub_url(repo_id=repo_id, filename=img_file.path, repo_type="dataset")
|
35 |
+
label_url = hf_hub_url(repo_id=repo_id, filename=label_path, repo_type="dataset")
|
36 |
+
|
37 |
+
local_image_path = Path(output_dir) / img_file.path
|
38 |
+
local_label_path = Path(output_dir) / label_path
|
39 |
+
|
40 |
+
local_image_path.parent.mkdir(parents=True, exist_ok=True)
|
41 |
+
local_label_path.parent.mkdir(parents=True, exist_ok=True)
|
42 |
+
|
43 |
+
try:
|
44 |
+
# Download and save the image
|
45 |
+
with fsspec.open(image_url) as f:
|
46 |
+
image = Image.open(BytesIO(f.read()))
|
47 |
+
image.save(local_image_path)
|
48 |
+
|
49 |
+
# Download and save the corresponding .txt label
|
50 |
+
with fsspec.open(label_url) as f:
|
51 |
+
txt_content = f.read()
|
52 |
+
with open(local_label_path, "wb") as out_f:
|
53 |
+
out_f.write(txt_content)
|
54 |
+
|
55 |
+
# print(f"[{count+1}] {rel_path} and {rel_path.with_suffix('.txt')}")
|
56 |
+
count += 1
|
57 |
+
except Exception as e:
|
58 |
+
print(f" Failed {rel_path}: {e}")
|
59 |
+
|
60 |
+
if count >= max_files:
|
61 |
+
break
|
62 |
+
|
63 |
+
print(f" Downloaded {count} image/txt pairs.")
|
64 |
+
print(f" Saved under: {Path(output_dir).resolve()}")
|
65 |
+
|
66 |
+
def parse_args():
|
67 |
+
parser = argparse.ArgumentParser(description="Stream and sample paired images + txt labels from a Hugging Face folder-structured dataset.")
|
68 |
+
parser.add_argument("--repo-id", required=False, default = "JeffreyJsam/SWiM-SpacecraftWithMasks",help="Hugging Face dataset repo ID.")
|
69 |
+
parser.add_argument("--image-subdir", required=False, default = "Baseline/images/val/000", help="Subdirectory path for images.")
|
70 |
+
parser.add_argument("--label-subdir", required=False, default="Baseline/labels/val/000", help="Subdirectory path for txt masks.")
|
71 |
+
parser.add_argument("--output-dir", default="./Sampled-SWiM", help="Where to save sampled data.")
|
72 |
+
parser.add_argument("--count", type=int, default=500, help="How many samples to download.")
|
73 |
+
return parser.parse_args()
|
74 |
+
|
75 |
+
if __name__ == "__main__":
|
76 |
+
args = parse_args()
|
77 |
+
sample_dataset(
|
78 |
+
repo_id=args.repo_id,
|
79 |
+
image_subdir=args.image_subdir,
|
80 |
+
label_subdir=args.label_subdir,
|
81 |
+
output_dir=args.output_dir,
|
82 |
+
max_files=args.count,
|
83 |
+
)
|