File size: 7,556 Bytes
094d5f3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
"""
export.py — Convert COCO annotations to YOLO training format.

The final step before training: takes the pipeline's COCO JSON output
and produces a YOLO-ready dataset with data.yaml.

Usage:
    # Convert a pipeline experiment to YOLO format
    data_label_factory export --experiment experiments/latest/ --output yolo_dataset/

    # Or specify a COCO file directly
    data_label_factory export --coco path/to/stop-signs.coco.json --images ~/data-label-factory/stop-signs --output yolo_dataset/
"""

from __future__ import annotations

import json
import os
import shutil
import random
from pathlib import Path


def coco_to_yolo(
    coco_path: str,
    image_root: str,
    output_dir: str,
    val_split: float = 0.1,
    copy_images: bool = True,
) -> dict:
    """Convert COCO annotations to YOLO format.

    Args:
        coco_path: Path to COCO JSON file
        image_root: Root directory where images live (file_name in COCO is relative to this)
        output_dir: Output directory for YOLO dataset
        val_split: Fraction of images for validation (default 0.1)
        copy_images: Whether to copy images to output dir (default True)

    Returns:
        Summary dict with paths and stats
    """
    with open(coco_path) as f:
        coco = json.load(f)

    images = {img["id"]: img for img in coco.get("images", [])}
    annotations = coco.get("annotations", [])
    categories = coco.get("categories", [])

    # Build category mapping: COCO cat_id → YOLO class_id (0-indexed)
    cat_id_to_yolo = {}
    cat_names = {}
    for i, cat in enumerate(categories):
        cat_id_to_yolo[cat["id"]] = i
        cat_names[i] = cat["name"]

    # Group annotations by image
    anns_by_image = {}
    for ann in annotations:
        anns_by_image.setdefault(ann["image_id"], []).append(ann)

    # Create output dirs
    out = Path(output_dir)
    train_img = out / "images" / "train"
    train_lbl = out / "labels" / "train"
    val_img = out / "images" / "val"
    val_lbl = out / "labels" / "val"
    for d in [train_img, train_lbl, val_img, val_lbl]:
        d.mkdir(parents=True, exist_ok=True)

    # Split images
    img_ids = list(images.keys())
    random.shuffle(img_ids)
    n_val = max(1, int(len(img_ids) * val_split))
    val_ids = set(img_ids[:n_val])
    train_ids = set(img_ids[n_val:])

    stats = {"train": 0, "val": 0, "annotations": 0, "skipped": 0}

    for img_id, img_info in images.items():
        is_val = img_id in val_ids
        img_dir = val_img if is_val else train_img
        lbl_dir = val_lbl if is_val else train_lbl

        iw = img_info.get("width", 1)
        ih = img_info.get("height", 1)
        fname = img_info.get("file_name", "")
        src_path = os.path.join(image_root, fname)

        if not os.path.exists(src_path):
            stats["skipped"] += 1
            continue

        # Copy image
        ext = os.path.splitext(fname)[1] or ".jpg"
        safe_name = fname.replace("/", "_").replace("\\", "_")
        dst_img = img_dir / safe_name
        if copy_images:
            shutil.copy2(src_path, dst_img)

        # Write YOLO label file
        stem = os.path.splitext(safe_name)[0]
        label_lines = []
        for ann in anns_by_image.get(img_id, []):
            cls_id = cat_id_to_yolo.get(ann.get("category_id"), 0)
            x, y, w, h = ann["bbox"]  # COCO: [x, y, w, h] in pixels
            # Convert to YOLO: [cx, cy, w, h] normalized 0-1
            cx = (x + w / 2) / iw
            cy = (y + h / 2) / ih
            nw = w / iw
            nh = h / ih
            # Clamp to [0, 1]
            cx = max(0, min(1, cx))
            cy = max(0, min(1, cy))
            nw = max(0, min(1, nw))
            nh = max(0, min(1, nh))
            label_lines.append(f"{cls_id} {cx:.6f} {cy:.6f} {nw:.6f} {nh:.6f}")
            stats["annotations"] += 1

        with open(lbl_dir / f"{stem}.txt", "w") as f:
            f.write("\n".join(label_lines))

        if is_val:
            stats["val"] += 1
        else:
            stats["train"] += 1

    # Write data.yaml
    import yaml
    data_yaml = {
        "path": str(out.resolve()),
        "train": "images/train",
        "val": "images/val",
        "nc": len(cat_names),
        "names": cat_names,
    }
    yaml_path = out / "data.yaml"
    with open(yaml_path, "w") as f:
        yaml.dump(data_yaml, f, default_flow_style=False)

    summary = {
        "output_dir": str(out),
        "data_yaml": str(yaml_path),
        "train_images": stats["train"],
        "val_images": stats["val"],
        "total_annotations": stats["annotations"],
        "skipped": stats["skipped"],
        "classes": cat_names,
        "nc": len(cat_names),
    }

    print(f"  YOLO dataset: {out}")
    print(f"  Train: {stats['train']} images, Val: {stats['val']} images")
    print(f"  Annotations: {stats['annotations']}")
    print(f"  Classes ({len(cat_names)}): {list(cat_names.values())}")
    print(f"  data.yaml: {yaml_path}")

    # Print training command
    print(f"\n  Training command:")
    print(f"    yolo detect train \\")
    print(f"      model=yolo11n.pt \\")
    print(f"      data={yaml_path} \\")
    print(f"      epochs=50 imgsz=640 batch=16 \\")
    print(f"      project=runs name={out.name}")

    return summary


def main(argv=None):
    import argparse
    p = argparse.ArgumentParser(
        prog="data_label_factory export",
        description="Convert COCO annotations to YOLO training format.",
    )
    p.add_argument("--coco", help="Path to COCO JSON file")
    p.add_argument("--experiment", help="Experiment directory (auto-finds COCO file)")
    p.add_argument("--images", help="Image root directory")
    p.add_argument("--output", default="yolo_dataset", help="Output directory")
    p.add_argument("--val-split", type=float, default=0.1, help="Validation split (default 0.1)")
    p.add_argument("--no-copy", action="store_true", help="Don't copy images (symlink instead)")
    args = p.parse_args(argv)

    coco_path = args.coco
    image_root = args.images

    if not coco_path and args.experiment:
        # Find COCO file in experiment
        exp_dir = args.experiment
        if exp_dir == "latest":
            from .experiments import list_experiments
            exps = list_experiments()
            if exps:
                exp_dir = exps[0]["path"]
        for dirpath, _, filenames in os.walk(exp_dir):
            for fn in filenames:
                if fn.endswith(".coco.json"):
                    coco_path = os.path.join(dirpath, fn)
                    break
        if not coco_path:
            print(f"No COCO file found in {exp_dir}")
            return

    if not coco_path:
        p.error("--coco or --experiment required")

    if not image_root:
        # Try to guess from COCO info
        with open(coco_path) as f:
            coco = json.load(f)
        target = coco.get("info", {}).get("target_object", "")
        project = coco.get("info", {}).get("description", "").split("for ")[-1].split(" via")[0]
        image_root = os.path.expanduser(f"~/data-label-factory/{project}")
        if not os.path.exists(image_root):
            print(f"  Image root not found: {image_root}")
            print(f"  Specify with --images")
            return

    print(f"Converting COCO → YOLO")
    print(f"  COCO: {coco_path}")
    print(f"  Images: {image_root}")
    print(f"  Output: {args.output}")
    coco_to_yolo(coco_path, image_root, args.output,
                 val_split=args.val_split, copy_images=not args.no_copy)