Pengyu Zhang

@DZSYUNNAN The update layer can be set in train_ADRNet.py line249 (ft_layers) and the custom dataset path can be added in line 36 with the instruction of attribute-specific dataset generation. In...

Hello, this seems an encoding type error[1,2], just try adding ``` #!/usr/bin/env python # coding=utf-8 ``` to the begining of the code. [1] https://blog.csdn.net/weixin_41529093/article/details/111343308 [2] https://blog.csdn.net/qq_36387683/article/details/91374774 If the same error...

Please refer to #8 to recompile the Precise ROI pooling

抱歉，因为我在训练的过程中没有参考IoU的值，这部分我没有实现，可以参考Precision()，写一个计算IoU的函数，输出训练过程中的IoU值

所有学习率都是一样的，需要把学习率调小至1e-5, 我在训练的时候没看到下降的情况，之前的层freeze了嘛

对的，但达到最好效果的模型可能不是最后一个iteration，你可以测试下早些iteration的效果

Thanks for your attention. The main question seems that if we can use concat + 1 fc layer to replace 5 fc layers. Theoretically, we aim to achieve attribute selection...

Sorry for late response. I can't understand how to use sigmoid and 1 fc to act as SENet. Can you provide more detailed description?

对的，我们想要让各自的selection部分的全连接层学习到各自分支对应的权重，可能和SENet的操作有一定区别，可以参考SKNet[Selective Kernel Networks, CVPR 2019]可能能更好的理解～

你好，训练代码已开源（详见$root/mfDiMP/ltr/run_training.py）,数据集已公开（https://github.com/zhang-pengyu/DUT-VTUAV）