计算机视觉
文章平均质量分 80
grllery
这个作者很懒,什么都没留下…
展开
-
《Swin Transformer》运算Graph简示
Swin Transformer 运算流程图简例Graph原创 2021-08-28 22:31:37 · 262 阅读 · 0 评论 -
论文简读《An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale》
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale (ViT)原创 2021-05-23 16:40:07 · 424 阅读 · 0 评论 -
论文简读《Harmonizing Transferability and Discriminability for Adapting Object Detectors》
CVPR2020 | Code思想:首先文章提出当前基于对抗的方法image and instance levels alignment [7], strong-local and weak-global alignment [44], local-region alignment based on region proposal [62], multi-level feature alignment with prediction-guided instance-level constraint原创 2021-03-14 22:01:55 · 855 阅读 · 2 评论 -
论文简读《Cross-domain Object Detection through Coarse-to-Fine Feature Adaptation》
CVPR 2020思想:强调前景的重要性,在对齐时提高判别器对前景部分的权重(RPN);除了类别不可知的对齐,引入细类别的对齐(Prototype)。Attentionbased Region Transfer (ART)首先对于层lll的对抗损失如下:LADVl=minθGlmaxθDlExs∈DSlog(Dl(Gl(xs)))+Ext∈DTlog(1−Dl(Gl(xt)))% \begin{equation} \begin{aligned} \mathcal{原创 2021-03-07 20:57:08 · 403 阅读 · 0 评论 -
论文简读《RepVGG: Making VGG-style ConvNets Great Again》
多分支的结构往往比单分支结构的性能要好,可能原因是`multi-branch`模型为众多浅层模型的隐式集合。因此作者提出在训练的时候使用`multi-branch`的结构,但是在部署的时候转换为单分支(不改变精度的情况下)。原创 2021-01-15 21:13:06 · 820 阅读 · 2 评论 -
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]. Analogou原创 2021-01-10 10:22:32 · 248 阅读 · 0 评论 -
沃尔什变换
沃尔什变换转载 2019-04-06 15:39:10 · 9510 阅读 · 0 评论