Code confusion

[zhangbububu]

t_grad_cam, s_grad_cam ===> (n_classes, d_model)

and t_grad_cam, s_grad_cam are inputs of channel_similarity

so why we have :

    bsz, ch = f_s.shape[0], f_s.shape[1]
    # Reshape
    f_s = f_s.view(bsz, ch, -1)
    f_t = f_t.view(bsz, ch, -1)

[IronMan]

Many thanks for your attention, our full code has been published! If you have any other questions, feel free to reach out by email, thank you!

Best
Lujun