目标检测与 YOLO 系列工程落地

import numpy as np

def nms(boxes, scores, iou_threshold=0.5):
    order = scores.argsort()[::-1]
    keep = []
    while order.size > 0:
        i = order[0]
        keep.append(int(i))
        if order.size == 1:
            break
        rest = order[1:]
        xx1 = np.maximum(boxes[i, 0], boxes[rest, 0])
        yy1 = np.maximum(boxes[i, 1], boxes[rest, 1])
        xx2 = np.minimum(boxes[i, 2], boxes[rest, 2])
        yy2 = np.minimum(boxes[i, 3], boxes[rest, 3])
        w = np.maximum(0.0, xx2 - xx1)
        h = np.maximum(0.0, yy2 - yy1)
        inter = w * h
        area_i = (boxes[i, 2] - boxes[i, 0]) * (boxes[i, 3] - boxes[i, 1])
        area_r = (boxes[rest, 2] - boxes[rest, 0]) * (boxes[rest, 3] - boxes[rest, 1])
        iou = inter / (area_i + area_r - inter + 1e-6)
        order = rest[iou <= iou_threshold]
    return keep

from ultralytics import YOLO

model = YOLO("yolov8n.pt")   # 自动下载 COCO 预训练权重

# 推理：conf 是置信度阈值，iou 是 NMS 阈值
results = model("street.jpg", conf=0.25, iou=0.45)[0]

for box in results.boxes:
    cls_id = int(box.cls)
    conf = float(box.conf)
    xyxy = box.xyxy[0].tolist()
    print(results.names[cls_id], f"{conf:.3f}", [round(v, 1) for v in xyxy])

print("detected:", len(results.boxes))

import numpy as np
import torch

def nms_torch(boxes, scores, iou_threshold=0.5):
    keep = nms(boxes.numpy(), scores.numpy(), iou_threshold)
    return torch.tensor(keep, dtype=torch.long)

boxes = torch.tensor([[0, 0, 10, 10], [1, 1, 11, 11], [50, 50, 60, 60]], dtype=torch.float32)
scores = torch.tensor([0.9, 0.8, 0.7])
print("keep:", nms_torch(boxes, scores, 0.5))   # 保留框 0 与 2，框 1 被抑制

# 逐类 NMS：多类别时必须按类别分别做，否则不同类别会互相抑制
def multiclass_nms(boxes, scores, labels, iou_thr=0.5):
    keep_all = []
    for c in labels.unique():
        m = labels == c
        keep_all += [int(i) for i in nms_torch(boxes[m], scores[m], iou_thr)]
    return keep_all

print(multiclass_nms(boxes, scores, torch.tensor([0, 0, 1])))

from ultralytics import YOLO

model = YOLO("yolov8n.pt")
model.train(
    data="dataset.yaml",   # 指定 train/val 路径与类别名
    epochs=100,
    imgsz=640,
    batch=16,
    lr0=0.01,
    patience=20,           # 早停
    mosaic=1.0,
    device=0,
)
metrics = model.val()
print(metrics.box.map, metrics.box.map50)

from ultralytics import YOLO

model = YOLO("runs/detect/train/weights/best.pt")
model.export(format="onnx", imgsz=640, opset=12, simplify=True)
model.export(format="engine", half=True)   # TensorRT FP16，需 GPU

import json
from pathlib import Path

def coco_to_yolo(coco_json, out_dir, img_dir):
    data = json.loads(Path(coco_json).read_text())
    images = {im["id"]: im for im in data["images"]}
    cats = {c["id"]: i for i, c in enumerate(data["categories"])}
    out = Path(out_dir)
    out.mkdir(parents=True, exist_ok=True)

    for ann in data["annotations"]:
        im = images[ann["image_id"]]
        W, H = im["width"], im["height"]
        x, y, w, h = ann["bbox"]                       # COCO 是左上角加宽高
        cx, cy = (x + w / 2) / W, (y + h / 2) / H      # 转归一化中心点
        line = f"{cats[ann['category_id']]} {cx:.6f} {cy:.6f} {w/W:.6f} {h/H:.6f}"
        with open(out / (Path(im["file_name"]).stem + ".txt"), "a") as f:
            f.write(line + "\n")

coco_to_yolo("annotations/instances_train.json", "labels/train", "images/train")

from ultralytics import YOLO

# 实例分割
seg = YOLO("yolov8n-seg.pt")
r = seg("street.jpg")[0]
print("masks:", None if r.masks is None else r.masks.data.shape)

# 姿态估计
pose = YOLO("yolov8n-pose.pt")
rp = pose("people.jpg")[0]
print("keypoints:", None if rp.keypoints is None else rp.keypoints.data.shape)

# 多目标跟踪：persist=True 在视频帧间保持轨迹
track = YOLO("yolov8n.pt")
for frame in ["f1.jpg", "f2.jpg", "f3.jpg"]:
    res = track.track(frame, persist=True)[0]
    ids = res.boxes.id
    print("track ids:", None if ids is None else ids.tolist())