Confused about the dim of feature size

[kkk123698745]

i ‘m confused about the dim of the output of the forward_feature. According to the paper, the input of the forward_up_features should be (b, d/4, h/32,w/32,8C) but the output of forward_feature is (b, h/32, h/32,w/32,8C). and i notice the function forward_features_to_token_learner are not used, so is there something wrong with the forward_features_to_token_learner.