search

Lliu666 / MSHNet

Implement of CVPR2024 'Infrared Small Target Detection with Scale and Location Sensitivity'
41 stars 10 forks source link

关于论文中decoder后的4个3*3的卷积和代码不对应，代码是1*1，请问是画错了还是代码有误 #6

Open GanGeeee opened 6 months ago

GanGeeee commented 6 months ago
截屏2024-04-24 14 50 12 截屏2024-04-24 14 38 35
Lliu666 commented 6 months ago

Sorry, this is our mistake. We drew the wrong picture in the paper. However, we re-used 3x3 convolution to conduct relevant experiments and found that kernel size does not have much impact on the final result. Here are the results under 3x3 convolution for your reference: IoU: 67.22, Pd: 94.56, Fa:14.88.