Qiang Hu

error

1

one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.FloatTensor [2, 128, 80, 88]], which is output 0 of ReluBackward0, is at version 3;...

question

1

请问boxlevelset在mmdet框架中测试，如何只输出掩膜图？就是不需要有原图的叠加

Detection and Localization

1

请问您在文章中，这两个指标的P,R,F1分别是怎么计算的？PraNet等是分割模型，是如何计算Detection的P,R,F1的？

It is well-established that when employing CNNs as segmentation backbones, multi-scale features are typically leveraged. However, when using ViTs as segmentation backbones, do we exclusively utilize the final layer's patch...

Motivation

1

Hi! I am troubled by the text in the red box (see in the figure) and I cannot understand why these two factors (i.e., the context information presented in different...