随风i的博客

多个类别分割mIoU的实现# 多类别miou计算方法def get_miou(outputs, masks, num_classes=20): # 计算 mIoU # 获取每个像素的预测类别，形状为 [batch_size, width, height] pred_labels = torch.argmax(outputs, dim=1) # 预测类别 total_miou = 0.0 # 初始化总 mIoU valid_class_count = 0 #