Datasets:

Modalities:
Image
Text
ArXiv:
License:
File size: 8,013 Bytes
82e3506
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
#!/usr/bin/env python3
"""
build_dataset.py — reconstruct the gRef-CW image set locally.

gRef-CW ships only the ANNOTATIONS (``grefs(unc).json`` and ``instances.json``).
The underlying images are part of the CropAndWeed dataset (AIT) and, under its
non-commercial licence, **may not be redistributed**. This script therefore
obtains the images from the official CropAndWeed source and arranges them to
match the gRef-CW annotations.

Pipeline
--------
  1. (``--auto``) clone the official CropAndWeed repo and run its downloader
     (``python cnw/setup.py``), or point at an existing CropAndWeed checkout
     with ``--cnw-images``.
  2. index the downloaded CropAndWeed images by file name / stem.
  3. for every image referenced by ``instances.json``, link (or copy) it into
     ``<data-dir>/images/``.
  4. verify that every referenced image is present; report anything missing.

Usage
-----
  # Fully automatic: clone CropAndWeed, download images, then build.
  python scripts/build_dataset.py --data-dir data --auto

  # If you already have CropAndWeed images on disk:
  python scripts/build_dataset.py --data-dir data \
      --cnw-images /path/to/cropandweed-dataset/data/images

  # Copy instead of symlink (e.g. across filesystems):
  python scripts/build_dataset.py --data-dir data --cnw-images <dir> --copy

Notes
-----
* CropAndWeed image files are named ``ave_<session>_<id>`` / ``vwg_<session>_<id>``
  (see the CropAndWeed README). Matching is done on the file name, falling back
  to the stem (name without extension), so it is tolerant of ``.jpg``/``.png``.
* Stdlib only — no third-party dependencies.
* Licence details: see ``DATASHEET.md`` and ``LICENSE``.
"""

from __future__ import annotations

import argparse
import json
import os
import shutil
import subprocess
import sys
from pathlib import Path

CNW_REPO = "https://github.com/cropandweed/cropandweed-dataset.git"
IMAGE_EXTS = (".jpg", ".jpeg", ".png", ".tif", ".tiff", ".bmp")


def log(msg: str) -> None:
    print(f"[build_dataset] {msg}", flush=True)


def referenced_file_names(instances_json: Path) -> list[str]:
    """Return the list of image file names referenced by instances.json (COCO-style)."""
    with instances_json.open(encoding="utf-8") as f:
        data = json.load(f)
    images = data.get("images")
    if not images:
        sys.exit(f"ERROR: no 'images' array found in {instances_json}. "
                 "Confirm this is a COCO-style instances file.")
    names = []
    for img in images:
        name = img.get("file_name") or img.get("filename") or img.get("name")
        if not name:
            sys.exit(f"ERROR: an entry in {instances_json} has no file_name field: {img!r}")
        names.append(name)
    return names


def index_images(images_dir: Path) -> dict[str, Path]:
    """Map both full file name and stem -> path for every image under images_dir."""
    if not images_dir.is_dir():
        sys.exit(f"ERROR: CropAndWeed images directory not found: {images_dir}")
    index: dict[str, Path] = {}
    count = 0
    for path in images_dir.rglob("*"):
        if path.is_file() and path.suffix.lower() in IMAGE_EXTS:
            index.setdefault(path.name, path)
            index.setdefault(path.stem, path)
            count += 1
    log(f"indexed {count} image files under {images_dir}")
    if count == 0:
        sys.exit("ERROR: no image files found. Did the CropAndWeed download succeed? "
                 "Pass the correct --cnw-images directory.")
    return index


def resolve(name: str, index: dict[str, Path]) -> Path | None:
    """Find a CropAndWeed image for a referenced gRef-CW file name."""
    if name in index:
        return index[name]
    stem = Path(name).stem
    return index.get(stem)


def auto_download(workdir: Path) -> Path:
    """Clone CropAndWeed and run its downloader. Returns the images directory."""
    repo_dir = workdir / "cropandweed-dataset"
    if not repo_dir.exists():
        log(f"cloning CropAndWeed into {repo_dir} ...")
        subprocess.run(["git", "clone", "--depth", "1", CNW_REPO, str(repo_dir)], check=True)
    else:
        log(f"reusing existing CropAndWeed checkout at {repo_dir}")

    setup = repo_dir / "cnw" / "setup.py"
    if not setup.exists():
        sys.exit(f"ERROR: expected CropAndWeed downloader at {setup} but it is missing. "
                 "Check the CropAndWeed repo layout and use --cnw-images instead.")
    log("running CropAndWeed downloader (python cnw/setup.py) — this fetches the images ...")
    subprocess.run([sys.executable, "cnw/setup.py"], cwd=str(repo_dir), check=True)

    # CropAndWeed extracts images under data/images by convention; fall back to a search.
    candidate = repo_dir / "data" / "images"
    if candidate.is_dir() and any(candidate.rglob("*")):
        return candidate
    for sub in repo_dir.rglob("images"):
        if sub.is_dir() and any(p.suffix.lower() in IMAGE_EXTS for p in sub.rglob("*") if p.is_file()):
            return sub
    sys.exit("ERROR: CropAndWeed download finished but no images directory was found. "
             "Locate it manually and re-run with --cnw-images.")


def link_or_copy(src: Path, dst: Path, copy: bool) -> None:
    if dst.exists() or dst.is_symlink():
        dst.unlink()
    if copy:
        shutil.copy2(src, dst)
    else:
        try:
            os.symlink(os.path.relpath(src, dst.parent), dst)
        except OSError:
            shutil.copy2(src, dst)  # symlinks unsupported (e.g. some Windows setups)


def main() -> int:
    ap = argparse.ArgumentParser(description="Reconstruct the gRef-CW image set from CropAndWeed.")
    ap.add_argument("--data-dir", type=Path, default=Path("data"),
                    help="gRef-CW data dir containing instances.json; images written to <data-dir>/images.")
    ap.add_argument("--instances-json", type=Path, default=None,
                    help="Path to instances.json (default: <data-dir>/instances.json).")
    src = ap.add_mutually_exclusive_group(required=True)
    src.add_argument("--auto", action="store_true",
                     help="Clone CropAndWeed and run its official downloader automatically.")
    src.add_argument("--cnw-images", type=Path,
                     help="Path to an existing CropAndWeed images directory.")
    ap.add_argument("--workdir", type=Path, default=Path(".cropandweed_cache"),
                    help="Where to clone CropAndWeed when using --auto.")
    ap.add_argument("--copy", action="store_true",
                    help="Copy image files instead of symlinking.")
    args = ap.parse_args()

    instances_json = args.instances_json or (args.data_dir / "instances.json")
    if not instances_json.exists():
        sys.exit(f"ERROR: {instances_json} not found. Download the gRef-CW annotations first "
                 "(see README 'Data and Weights').")

    out_dir = args.data_dir / "images"
    out_dir.mkdir(parents=True, exist_ok=True)

    cnw_images = auto_download(args.workdir) if args.auto else args.cnw_images
    log(f"CropAndWeed images: {cnw_images}")

    index = index_images(cnw_images)
    names = referenced_file_names(instances_json)
    log(f"{len(names)} images referenced by {instances_json.name}")

    matched, missing = 0, []
    for name in names:
        src_path = resolve(name, index)
        if src_path is None:
            missing.append(name)
            continue
        link_or_copy(src_path, out_dir / Path(name).name, args.copy)
        matched += 1

    log(f"linked/copied {matched}/{len(names)} images into {out_dir}")
    if missing:
        log(f"WARNING: {len(missing)} referenced images were NOT found in CropAndWeed. "
            f"First few: {missing[:10]}")
        log("This usually means the CropAndWeed download is incomplete or the file-name "
            "convention differs — verify the source and re-run.")
        return 1
    log("Done — every referenced image was found. Dataset is ready.")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())