yolo v5召回率精确率代码

以下是使用 YOLOv5 模型计算召回率和精确率的示例代码：

import torch
from utils.general import non_max_suppression

# 加载模型和数据
model = torch.hub.load('ultralytics/yolov5', 'custom', path_or_model='path/to/model.pt')
dataset = torch.utils.data.DataLoader('path/to/dataset')

# 设置阈值和IOU阈值
conf_threshold = 0.5
iou_threshold = 0.5

# 计算召回率和精确率
total_true_positives = 0
total_false_positives = 0
total_false_negatives = 0
for images, targets in dataset:
    # 推理
    results = model(images)
    # 非极大值抑制
    results = non_max_suppression(results, conf_threshold, iou_threshold)
    # 计算真阳性和假阳性的数目
    true_positives = 0
    false_positives = 0
    for i, (result, target) in enumerate(zip(results, targets)):
        if result is None:
            total_false_negatives += 1
            continue
        # 计算IOU
        ious = torch.bbox_iou(result[:, :4], target[:, :4])
        # 找到IOU最大的预测框
        max_iou, max_iou_idx = ious.max(1)
        # 如果IOU大于阈值，则为真阳性；否则为假阳性
        true_positives += (max_iou >= iou_threshold).sum().item()
        false_positives += (max_iou < iou_threshold).sum().item()
    total_true_positives += true_positives
    total_false_positives += false_positives

# 计算精确率和召回率
precision = total_true_positives / (total_true_positives + total_false_positives)
recall = total_true_positives / (total_true_positives + total_false_negatives)
print('Precision: {:.4f}, Recall: {:.4f}'.format(precision, recall))