if there are more than 3 channels ,how does using "imagenet" as encoder weights work?

[Sumana18]

I am currently developing a model using Sentinel data with 11 bands. Since ImageNet weights are trained on 3-channel RGB images, I’m trying to understand how the following code is able to work with my 11-band input. Could you please explain how this is handled?

geoai.train_segmentation_model(
    images_dir=f"{data_folder}/images",
    labels_dir=f"{data_folder}/labels",
    output_dir=f"{data_folder}/unet_models",
    architecture="unet",
    encoder_name="resnet50",
    encoder_weights="imagenet",
    num_channels=11,  # Sentinel has 11 bands
    num_classes=2, 
    batch_size=8,
    num_epochs=100,
    learning_rate=0.0001,
    val_split=0.3,
    seed=42,
    verbose=True,
    early_stopping_patience=10,
)

[giswqs]

It builds upon pytorch segmentation models, which can handle more than 3 channels. See https://smp.readthedocs.io/en/latest