daifeng2016
daifeng2016
Got it, thank you.
Well, 0.5 is the simple threshold, or else you can use other binary segmentation algorithms such as ostu.
Hi， all your input to the net should be 4-D tensor (Batchsize, W,H,C), maybe your input of 'train_label' is not.
shifted and scaled is also a transformation, which is listed in our article.
Hi, it is quite easy to devise a dataloader for training and testing using keras. As our project are embedded in a large project, we have not yet build a...
Hi, the output of each scale (in our article it is 4) are fused to generate a new output, then a sigmoid layer is followed. In total, five outputs are...
Hi, as for your questions: 1) the kernel number of each convolution layer used in our paper is different from the original literature, for the high complexity of satellite images....
Please read our paper carefully, they are all included.
Hi，I use model.fit_generator function to process the data, which is commonly used when dealing with large training data. Please follow the example on the Internet for the usage of the...