提示：文章写完后，目录可以自动生成，如何生成可参考右边的帮助文档

---

前言

本文主要用于记录数据标注和模型预测之间的漏报和误报思想及其源码

---

提示：以下是本篇文章正文内容，下面案例可供参考

一、误报

我自己定义的误报是模型的预测结果框比人为标注的目标框多，也就是当标注人员标注图片的时候标注不仔细未能标注全的情况，逻辑是将在原始标注的xml文件当中添加误报-类别名称的框。

二、漏报

我自己定义的漏报是人为标注的框模型没有全部预测出来，也就是当标注人员标注图片的时候标注错误或者标注的框质量不合格的情况（跟模型性能也有关系），逻辑是将在原始标注的xml文件当中添加漏报-类别名称的框。

三、源码

import argparse
import os
import time
import shutil
import cv2
import numpy as np
import torch
from pathlib import Path
from pascal_voc_writer import Writer
import torchvision
from xml.etree import ElementTree
from xml.etree.ElementTree import Element

import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)

FILE = Path(__file__).resolve()
ROOT = FILE.parents[0]


def xywh2xyxy(x):
    # Convert nx4 boxes from [x, y, w, h] to [x1, y1, x2, y2] where xy1=top-left, xy2=bottom-right
    y = x.clone() if isinstance(x, torch.Tensor) else np.copy(x)
    y[:, 0] = x[:, 0] - x[:, 2] / 2  # top left x
    y[:, 1] = x[:, 1] - x[:, 3] / 2  # top left y
    y[:, 2] = x[:, 0] + x[:, 2] / 2  # bottom right x
    y[:, 3] = x[:, 1] + x[:, 3] / 2  # bottom right y
    return y


def box_iou(box1, box2):

    def box_area(box):
        # box = 4xn
        return (box[2] - box[0]) * (box[3] - box[1])

    area1 = box_area(box1.T)
    area2 = box_area(box2.T)

    # inter(N,M) = (rb(N,M,2) - lt(N,M,2)).clamp(0).prod(2)
    inter = (torch.min(box1[:, None, 2:], box2[:, 2:]) - torch.max(box1[:, None, :2], box2[:, :2])).clamp(0).prod(2)
    return inter / (area1[:, None] + area2 - inter)  # iou = inter / (area1 + area2 - inter)

def cv_imread(file_path):
    cv_img = cv2.imdecode(np.fromfile(file_path, dtype=np.uint8), cv2.IMREAD_UNCHANGED) #读取的为bgr图像
    return cv2.cvtColor(cv_img, cv2.COLOR_BGR2RGB)

def letterbox(im, new_shape=(640, 640), color=(114, 114, 114), auto=True, scaleFill=False, scaleup=True, stride=32):
    # Resize and pad image while meeting stride-multiple constraints
    shape = im.shape[:2]  # current shape [height, width]
    if isinstance(new_shape, int):
        new_shape = (new_shape, new_shape)

    # Scale ratio (new / old)
    r = min(new_shape[0] / shape[0], new_shape[1] / shape[1])
    if not scaleup:  # only scale down, do not scale up (for better val mAP)
        r = min(r, 1.0)

    # Compute padding
    ratio = r, r  # width, height ratios
    new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r))
    dw, dh = new_shape[1] - new_unpad[0], new_shape[0] - new_unpad[1]  # wh padding
    if auto:  # minimum rectangle
        dw, dh = np.mod(dw, stride), np.mod(dh, stride)  # wh padding
    elif scaleFill:  # stretch
        dw, dh = 0.0, 0.0
        new_unpad = (new_shape[1], new_shape[0])
        ratio = new_shape[1] / shape[1], new_shape[0] / shape[0]  # width, height ratios

    dw /= 2  # divide padding into 2 sides
    dh /= 2

    if shape[::-1] != new_unpad:  # resize
        im = cv2.resize(im, new_unpad, interpolation=cv2.INTER_LINEAR)
    top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1))
    left, right = int(round(dw - 0.1)), int(round(dw + 0.1))
    im = cv2.copyMakeBorder(im, top, bottom, left, right, cv2.BORDER_CONSTANT, value=color)  # add border
    return im, ratio, (dw, dh)

def preprocess_file(path, img_size, stride, auto):
    img_rgb_ = cv_imread(path)  # RGB
    assert img_rgb_ is not None, f'Image Not Found {path}'
    # Padded resize
    img_rgb = letterbox(img_rgb_, img_size, stride=stride, auto=auto)[0]

    # Convert
    img_rgb = img_rgb.transpose((2, 0, 1))  # HWC to CHW
    img_rgb = np.ascontiguousarray(img_rgb)# 将一个内存不连续存储的数组转换为内存连续存储的数组，使得运行速度更快
    return img_rgb, img_rgb_


def preprocess_mat(mat, img_size, stride, auto):
    img_bgr = mat  # BGR
    # Padded resize
    img_rgb = letterbox(img_bgr, img_size, stride=stride, auto=auto)[0]

    # Convert
    img_rgb = img_rgb.transpose((2, 0, 1))[::-1]  # HWC to CHW, BGR to RGB
    img_rgb = np.ascontiguousarray(img_rgb)
    return img_rgb, img_bgr


def clip_coords(boxes, shape):
    # Clip bounding xyxy bounding boxes to image shape (height, width)
    if isinstance(boxes, torch.Tensor):  # faster individually
        boxes[:, 0].clamp_(0, shape[1])  # x1
        boxes[:, 1].clamp_(0, shape[0])  # y1
        boxes[:, 2].clamp_(0, shape[1])  # x2
        boxes[:, 3].clamp_(0, shape[0])  # y2
    else:  # np.array (faster grouped)
        boxes[:, [0, 2]] = boxes[:, [0, 2]].clip(0, shape[1])  # x1, x2
        boxes[:, [1, 3]] = boxes[:, [1, 3]].clip(0, shape[0])  # y1, y2


def scale_coords(img1_shape, coords, img0_shape, ratio_pad=None):
    # Rescale coords (xyxy) from img1_shape to img0_shape
    if ratio_pad is None:  # calculate from img0_shape
        gain = min(img1_shape[0] / img0_shape[0], img1_shape[1] / img0_shape[1])  # gain  = old / new
        pad = (img1_shape[1] - img0_shape[1] * gain) / 2, (img1_shape[0] - img0_shape[0] * gain) / 2  # wh padding
    else:
        gain = ratio_pad[0][0]
        pad = ratio_pad[1]

    coords[:, [0, 2]] -= pad[0]  # x padding
    coords[:, [1, 3]] -= pad[1]  # y padding
    coords[:, :4] /= gain
    clip_coords(coords, img0_shape)
    return coords

def remove_name_elements(element):
    name_element = element.find('name')
    if name_element is not None and name_element.text and name_element.text.startswith('\ufeff'):
        name_element.text = name_element.text.lstrip('\ufeff')
    for child in element:
        remove_name_elements(child)

def read_xml(xml_file: str, names):
    if os.path.getsize(xml_file) == 0:
        return []


    with open(xml_file, encoding='utf-8-sig') as in_file:
        # if not in_file.readline():
        #     return []
        tree = ElementTree.parse(in_file)
        root = tree.getroot()
    remove_name_elements(root)

    results = []
    obj: Element
    for obj in tree.findall("object"):
        xml_box = obj.find("bndbox")

        x_min = float(xml_box.find("xmin").text)
        y_min = float(xml_box.find("ymin").text)

        x_max = float(xml_box.find("xmax").text)
        y_max = float(xml_box.find("ymax").text)

        b = [x_min, y_min, x_max, y_max]
        cls_id = names.index(obj.find("name").text)
        results.append([cls_id, b])
    return results


def non_max_suppression(prediction, conf_thres=0.25, iou_thres=0.45, classes=None, agnostic=False, multi_label=False,
                        labels=(), max_det=300):
    """Runs Non-Maximum Suppression (NMS) on inference results
    Returns:
         list of detections, on (n,6) tensor per image [xyxy, conf, cls]
    """

    nc = prediction.shape[2] - 5  # number of classes
    xc = prediction[..., 4] > conf_thres  # candidates

    # Checks
    assert 0 <= conf_thres <= 1, f'Invalid Confidence threshold {conf_thres}, valid values are between 0.0 and 1.0'
    assert 0 <= iou_thres <= 1, f'Invalid IoU {iou_thres}, valid values are between 0.0 and 1.0'

    # Settings
    min_wh, max_wh = 2, 7680  # (pixels) minimum and maximum box width and height
    max_nms = 30000  # maximum number of boxes into torchvision.ops.nms()
    time_limit = 10.0  # seconds to quit after
    redundant = True  # require redundant detections
    multi_label &= nc > 1  # multiple labels per box (adds 0.5ms/img)
    merge = False  # use merge-NMS

    t = time.time()
    output = [torch.zeros((0, 6), device=prediction.device)] * prediction.shape[0]
    for xi, x in enumerate(prediction):  # image index, image inference
        # Apply constraints
        x[((x[..., 2:4] < min_wh) | (x[..., 2:4] > max_wh)).any(1), 4] = 0  # width-height
        x = x[xc[xi]]  # confidence

        # Cat apriori labels if autolabelling
        if labels and len(labels[xi]):
            lb = labels[xi]
            v = torch.zeros((len(lb), nc + 5), device=x.device)
            v[:, :4] = lb[:, 1:5]  # box
            v[:, 4] = 1.0  # conf
            v[range(len(lb)), lb[:, 0].long() + 5] = 1.0  # cls
            x = torch.cat((x, v), 0)

        # If none remain process next image
        if not x.shape[0]:
            continue

        # Compute conf
        x[:, 5:] *= x[:, 4:5]  # conf = obj_conf * cls_conf

        # Box (center x, center y, width, height) to (x1, y1, x2, y2)
        box = xywh2xyxy(x[:, :4])

        # Detections matrix nx6 (xyxy, conf, cls)
        if multi_label:
            i, j = (x[:, 5:] > conf_thres).nonzero(as_tuple=False).T
            x = torch.cat((box[i], x[i, j + 5, None], j[:, None].float()), 1)
        else:  # conf是置信度 j是类别
            conf, j = x[:, 5:].max(1, keepdim=True)
            x = torch.cat((box, conf, j.float()), 1)[conf.view(-1) > conf_thres]

        # Filter by class
        if classes is not None:
            x = x[(x[:, 5:6] == torch.tensor(classes, device=x.device)).any(1)]

        # Apply finite constraint
        # if not torch.isfinite(x).all():
        #     x = x[torch.isfinite(x).all(1)]

        # Check shape
        n = x.shape[0]  # number of boxes
        if not n:  # no boxes
            continue
        elif n > max_nms:  # excess boxes
            x = x[x[:, 4].argsort(descending=True)[:max_nms]]  # sort by confidence

        # Batched NMS
        c = x[:, 5:6] * (0 if agnostic else max_wh)  # classes
        boxes, scores = x[:, :4] + c, x[:, 4]  # boxes (offset by class), scores
        i = torchvision.ops.nms(boxes, scores, iou_thres)  # NMS
        if i.shape[0] > max_det:  # limit detections
            i = i[:max_det]
        if merge and (1 < n < 3E3):  # Merge NMS (boxes merged using weighted mean)
            # update boxes as boxes(i,4) = weights(i,n) * boxes(n,4)
            iou = box_iou(boxes[i], boxes) > iou_thres  # iou matrix
            weights = iou * scores[None]  # box weights
            x[i, :4] = torch.mm(weights, x[:, :4]).float() / weights.sum(1, keepdim=True)  # merged boxes
            if redundant:
                i = i[iou.sum(1) > 1]  # require redundancy

        output[xi] = x[i]
        if (time.time() - t) > time_limit:
            break  # time limit exceeded
    end = time.time()
    # print(time.time() - t,'seconds')
    return output

class Detect():
    def __init__(self, weights, imgsz, conf_thres, iou_thres):
        self.device = 'cpu'
        self.weights = weights
        self.model = None
        self.imgsz = imgsz
        self.conf_thres = conf_thres
        self.iou_thres = iou_thres
        if torch.cuda.is_available() and torch.cuda.device_count() > 1:
            self.device = torch.device('cuda:0')

        self.init_model()
        self.stride = max(int(self.model.stride.max()), 32)

    def init_model(self):
        ckpt = torch.load(self.weights, map_location=self.device)  # load
        ckpt = (ckpt.get('ema', None) or ckpt['model']).float()  # FP32 model
        fuse = True
        self.model = ckpt.fuse().eval() if fuse else ckpt.eval()  # fused or un-fused model in eval mode fuse()将Conv和bn层进行合并，提高模型的推理速度
        self.model.float()

    def infer_image(self, image_path):
        im, im0 = preprocess_file(image_path, img_size=self.imgsz, stride=self.stride, auto=True)
        im = torch.from_numpy(im).to(self.device).float() / 255
        if len(im.shape) == 3:
            im = im[None]  # expand for batch dim

        # Inference
        pred = self.model(im, augment=False, visualize=False)[0]
        # NMS
        pred = non_max_suppression(pred, self.conf_thres, self.iou_thres, None, False, max_det=1000)
        det = pred[0]

        results = []
        if len(det):
            # Rescale boxes from img_size to im0 size
            det[:, :4] = scale_coords(im.shape[2:], det[:, :4], im0.shape).round()
            # results
            for *xyxy, conf, cls in reversed(det):
                xyxy = (torch.tensor(xyxy).view(1, 4)).view(-1).tolist()  # normalized xywh
                results.append([cls.item(), xyxy, conf.item()])

        return results

    def infer_mat(self, mat):
        im, im0 = preprocess_mat(mat, img_size=self.imgsz, stride=self.stride, auto=True)
        im = torch.from_numpy(im).to(self.device).float() / 255
        if len(im.shape) == 3:
            im = im[None]  # expand for batch dim

        # Inference
        pred = self.model(im, augment=False, visualize=False)[0]
        # NMS
        pred = non_max_suppression(pred, self.conf_thres, self.iou_thres, None, False, max_det=1000)
        det = pred[0]

        results = []
        if len(det):
            # Rescale boxes from img_size to im0 size
            det[:, :4] = scale_coords(im.shape[2:], det[:, :4], im0.shape).round()
            # results
            for *xyxy, conf, cls in reversed(det):
                xyxy = (torch.tensor(xyxy).view(1, 4)).view(-1).tolist()  # normalized xywh
                results.append([cls.item(), xyxy, conf.item()])

        return results

def box_iou_np(box1, box2):

    x11, y11, x12, y12 = box1
    x21, y21, x22, y22 = box2
    width1 = np.maximum(0, x12 - x11)
    height1 = np.maximum(0, y12 - y11)
    width2 = np.maximum(0, x22 - x21)
    height2 = np.maximum(0, y22 - y21)
    area1 = width1 * height1
    area2 = width2 * height2
    # 计算交集，需要计算交集部分的左、上、右、下坐标
    xi1 = np.maximum(x11, x21)
    yi1 = np.maximum(y11, y21)
    xi2 = np.minimum(x12, x22)
    yi2 = np.minimum(y12, y22)
    # 计算交集部分面积
    w = np.maximum(0, xi2 - xi1)
    h = np.maximum(0, yi2 - yi1)
    intersection = w * h
    # 计算并集
    union = area1 + area2 - intersection
    # 计算iou
    iou = intersection / union
    return iou

def main(opt):
    if not os.path.exists(opt.output_path):
        os.makedirs(opt.output_path, exist_ok=True)#oxist_ok表示如果目录存在，不要抛出异常，正常结束
    detect = Detect(opt.weights, opt.imgsz, opt.conf_thres, opt.iou_thres)
    imgs = []
    for root,dirs,files in os.walk(opt.input_path):
        for file in files:
            if os.path.splitext(file)[1] in opt.extensions:
                imgs.append(root+'/'+file)
    total = len(imgs)
    for i,img in enumerate(imgs):
        print(f"{i + 1 : >05d}/{total : >05d} {img}")

        mat = cv_imread(img)
        xml = os.path.splitext(img)[0]+'.xml'
        h,w,_ = mat.shape
        results = detect.infer_image(img)

        # 标注
        anns = []
        if os.path.exists(xml):
            anns = read_xml(xml, opt.names)
        else:
            anns = []
        # 核查误报
        fps = []
        if opt.fp:
            for result in results:
                result_cls, result_box, _ = result
                if result_cls in opt.verifynames:
                    finded = False
                    for ann in anns:
                        ann_cls, ann_box = ann
                        if ann_cls == result_cls and box_iou_np(ann_box, result_box) > 0:
                            finded = True
                            break
                    if not finded:
                        fps.append([result_cls, result_box])
        # 核查漏报
        fns = []
        if opt.fn:
            for ann in anns:
                ann_cls, ann_box = ann
                if ann_cls in opt.verifynames:
                    finded = False
                    for result in results:
                        result_cls, result_box, _ = result
                        if ann_cls == result_cls and box_iou_np(ann_box, result_box) > 0:
                            finded = True
                            break
                    if not finded:
                        fns.append([ann_cls, ann_box])

        if len(fps) == 0 and len(fns) == 0:
            continue

        # 写文件
        writer = Writer(img, w, h)
        # 写原始标注
        for ann in anns:
            ann_cls, ann_box = ann
            x_min = ann_box[0]
            y_min = ann_box[1]
            x_max = ann_box[2]
            y_max = ann_box[3]
            writer.addObject(opt.names[int(ann_cls)], x_min, y_min, x_max, y_max)
        # 写误报
        if opt.fp:
            for ann in fps:
                ann_cls, ann_box = ann
                x_min = ann_box[0]
                y_min = ann_box[1]
                x_max = ann_box[2]
                y_max = ann_box[3]
                writer.addObject("误报-" + opt.names[int(ann_cls)], x_min, y_min, x_max, y_max)
        # 写漏报
        if opt.fn:
            for ann in fns:
                ann_cls, ann_box = ann
                x_min = ann_box[0]
                y_min = ann_box[1]
                x_max = ann_box[2]
                y_max = ann_box[3]
                writer.addObject("漏报-" + opt.names[int(ann_cls)], x_min, y_min, x_max, y_max)
        # 写文件
        writer.save(os.path.join(opt.output_path, os.path.basename(xml)))
        shutil.copy2(img, os.path.join(opt.output_path, os.path.basename(img)))


def parse_opt(known):
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights',type=str, default=ROOT / 'weights/best.pt', help='模型权重pt文件')
    parser.add_argument('--imgsz', type=tuple, default=(1280,1280), help='输入模型大小')
    parser.add_argument("--conf_thres", type=float, default=0.25, help="模型conf阈值")
    parser.add_argument('--iou_thres', type=float, default=0.5, help='标注与模型输出框的IOU阈值，用于判断误报和漏报')
    parser.add_argument('--names', type=list, default=["键盘", "显示器", "鼠标", "桌子", "椅子", "人"],help='核查的所有类别标注名称')
    parser.add_argument('--verifynames', type=list, default=[0,1], help='需要核查的类别')
    parser.add_argument('--input_path', type=str, default=r'', help='输入image和xml路径')
    parser.add_argument('--output_path', type=str, default=r''+'核查', help='输出image和xml路径')
    parser.add_argument('--extensions', type=list, default=['.jpg', '.JPG', '.jpeg', '.png', '.bmp', '.tiff', '.tif', '.svg', '.pfg'])
    parser.add_argument("--fp", type=bool, default=True, help="是否核查误报")
    parser.add_argument("--fn", type=bool, default=True, help="是否核查漏报")
    return parser.parse_known_args()[0] if known else parser.parse_args() #True 标志可以处理任何位置参数，不会因为位置参数崩溃，Fakse任何未知参数导致程序显示错误消息并退出

if __name__ == '__main__':
    opt = parse_opt(True)
    main(opt)

---

总结

安装对应的库，修改命令行参数weights、names、verifynames、input_path和output_path即可使用。（注：将源码放置到yolov5对应的文件夹下方即可。）