Buckets:
| """Render native-resolution bridge FN/FP crops from full-scene development records.""" | |
| import argparse | |
| import json | |
| from pathlib import Path | |
| import numpy as np | |
| from PIL import Image, ImageDraw | |
| from evaluation.locate_detection_errors import locate | |
| from training.model_factory import ROOT | |
| SCENES = ROOT / "datasets/external/dota-fullscene-dev/images" | |
| def draw(draw, points, colour, offset): | |
| vertices = [(x - offset[0], y - offset[1]) for x, y in zip(points[::2], points[1::2])] | |
| if vertices: | |
| draw.line(vertices + vertices[:1], fill=colour, width=4) | |
| def main() -> None: | |
| parser = argparse.ArgumentParser() | |
| parser.add_argument("records", type=Path) | |
| parser.add_argument("output", type=Path) | |
| parser.add_argument("--per-kind", type=int, default=20) | |
| parser.add_argument("--crop-size", type=int, default=1024) | |
| args = parser.parse_args() | |
| if args.output.exists(): | |
| raise FileExistsError("choose a fresh output directory") | |
| data = json.loads(args.records.read_text()) | |
| bridge = locate(data)["3"] | |
| targets = {row["sample_id"]: row["annotations"] for row in data["targets"]} | |
| predictions = {row["sample_id"]: row["annotations"] for row in data["predictions"]} | |
| args.output.mkdir(parents=True) | |
| manifest = {"status": "PENDING_SOURCE_LABEL_REVIEW", "class": "bridge", "records": str(args.records), | |
| "items": []} | |
| for kind in ("false_positives", "missed_targets"): | |
| for index, item in enumerate(bridge[kind][:args.per_kind]): | |
| scene = item["sample_id"] | |
| image_path = next(SCENES.rglob(scene + ".png")) | |
| image = Image.open(image_path).convert("RGB") | |
| points = np.asarray(item["obb_polygon"], dtype=float).reshape(-1, 2) | |
| cx, cy = points.mean(axis=0) | |
| left = max(0, min(round(cx - args.crop_size / 2), image.width - args.crop_size)) | |
| top = max(0, min(round(cy - args.crop_size / 2), image.height - args.crop_size)) | |
| crop = image.crop((left, top, left + args.crop_size, top + args.crop_size)) | |
| drawer = ImageDraw.Draw(crop) | |
| for target in targets[scene]: | |
| if target["category_id"] == 3: | |
| draw(drawer, target["obb_polygon"], "#25d366", (left, top)) | |
| for prediction in predictions[scene]: | |
| if prediction["category_id"] == 3: | |
| draw(drawer, prediction["obb_polygon"], "#f44336", (left, top)) | |
| draw(drawer, item["obb_polygon"], "#ffd600", (left, top)) | |
| name = f"bridge-{kind}-{index:02d}-{scene}.jpg" | |
| crop.save(args.output / name) | |
| manifest["items"].append({"kind": kind, "scene_id": scene, "file": name, | |
| "confidence": item.get("confidence"), "best_iou": item.get("best_iou"), | |
| "crop_box": [left, top, left + args.crop_size, top + args.crop_size], | |
| "review_status": "PENDING", "training_eligible": False}) | |
| (args.output / "manifest.json").write_text(json.dumps(manifest, indent=2) + "\n") | |
| print(json.dumps({"items": len(manifest["items"]), "output": str(args.output)})) | |
| if __name__ == "__main__": | |
| main() | |
Xet Storage Details
- Size:
- 3.24 kB
- Xet hash:
- 5c0b8507eb5c21671895c06c309a5691b146b2a026a43f09e4caf1c9211ae00b
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.