doubt regarding code

[CharveeKrishna]

RuntimeError: Given groups=1, weight of size [100, 128, 1, 1], expected input[9, 103, 9, 9] to have 128 channels, but got 103 channels instead.
This error pops up by running the DCFSL-UP.py file.
Can you please help resolve this issue.
Hoping for a reply soon and please treat this as emergency.
Thanks and Regards,
Charvee

[Li-ZK]

Hello, thank you for your question. I tried to check the code, so I downloaded the source code from GitHub again and followed the README.md file to download the data file Chikusei_ imdb_ 128. pickle and PaviaU.mat PaviaU_ gt. mat to the specified path, and the code can work properly. I tried to reproduce this error. According to the error prompt, the dimension does not correspond. The 128 dimension corresponds to the channel dimension of the Chikusei dataset in the source domain, and the 103 dimension corresponds to the channel dimension of the PaviaU dataset in the target domain. I suspect that the input data is PaviaU data in the 103 dimension. However, the mapping function called is for the 128->103 mapping of the source domain data, Following this idea, I removed domain='target 'from line 524, which would call the mapping function for the source domain and result in the error you mentioned.

support_ features, support_ outputs = feature_ encoder(supports.cuda(), domain='target') # This is code in 524

Of course, your mistake may not be the same as what I said. I suspect you have modified some experimental settings. I suggest checking if it is a feature_ The encoder is calling from the source domain, resulting in an incorrect dimension correspondence. You can also download the code again without modifying it, place the dataset in the correct location, and see if the code works properly.

Thanks and Regards,
Zhaokui Li