Train data with 4 channels

[kehuangke]

🚀 Feature

Can the tool train 4 channels of data?

Motivation

we construct our data into 3 channels and input it into TorchSSL model. We want to use more information to improve the discriminant accuracy of the training model. However, after we construct our data into 4 channels, the model give me an error.

Pitch

Alternatives

Additional context

[Hhhhhhao]

Hi, currently we only support CNNs on natural images with 3 channels:

Semi-supervised-learning/semilearn/nets/wrn/wrn.py

Line 84 in 0b73e35

self.conv1 = nn.Conv2d(3, channels[0], kernel_size=3, stride=1,

You can try to create you own network or replace the current one with 4 channels input.