Galileo encoder is never called in forward

[alecsandrei]

Hi,

Thanks for releasing the benchmark!
In galileo_model.py, I think the Galileo encoder is never called in forward (affects both GalileoClassifier and GalileoSegmentation):

self.embed_layer = backbone.space_time_embed[self.embed_group_name]
...
# Forward through the embedding layer
tokens = self.embed_layer(x)

self.embed_layer is a FlexiPatchEmbed, which only has a projection layer, Encoder.forward is never called (so no pass through the attention blocks). I'm not sure if this was intended.