论文简读《Exploring Categorical Regularization for Domain Adaptive Object Detection》

https://arxiv.org/pdf/2003.09152.pdf
提出类别正则化框架，主要使用多标签分类来进行实现前景物体的弱监督。

It is widely acknowledged that CNNs trained for singlelabel image classification tend to produce high responses on the local regions containing the main objects [38, 40, 39]. Analogously, CNNs trained for multi-label classification also have the weakly localization ability for the objects associated with image-level categories [35, 36].

并将图像级（Image Level）的多标签结果与实例级（Instance Level）的预测结果进行监督，挖掘目标域实例中的难样本（对实例赋予不同的损失权重）。

source code

        target_weight = []
        for i in range(len(tgt_pre_label)):
            label_i = tgt_pre_label[i].item()
            if label_i > 0:
                diff_value = torch.exp(
                    weight_value
                    * torch.abs(tgt_image_cls_feat[label_i - 1] - tgt_prob[i][label_i])
                ).item()
                target_weight.append(diff_value)
            else:
                target_weight.append(1.0)

        tgt_instance_loss = nn.BCELoss(
            weight=torch.Tensor(target_weight).view(-1, 1).cuda()
        )

最后

以上就是高贵未来为你收集整理的论文简读《Exploring Categorical Regularization for Domain Adaptive Object Detection》的全部内容，希望文章能够帮你解决论文简读《Exploring Categorical Regularization for Domain Adaptive Object Detection》所遇到的程序开发问题。

如果觉得靠谱客网站的内容还不错，欢迎将靠谱客网站推荐给程序员好友。