Bolin Ni
Bolin Ni
@notagenius have you solved this problem?
Hi, thanks for your interest in our work. We set it up like this because the output_dim should be the same as the input dimensions of the next transformer block.
Hi, thanks for your interest in our work. We write a function called [get_complexity](https://github.com/microsoft/Cream/blob/43514f073c64b804685aa9b3671ad77ef12b513d/AutoFormer/model/supernet_transformer.py#L295) to calculate the Flops. After you have set the subnet, you can use this function to...
Hi, thanks for your interest in our work. We directly use the searched model without retraining. For your question, it seems that your model is overfitting. The large model is...
Thanks for pointing out this mistake. We will fix it in a later revision.
Becauses the `DeformConv2D `has a problem, it only can do "SAME" conv. In `conv_kernel`, the shape of `x_offset ` returned from `_reshape_x_offset` is always (h*ks, w*ks), `ks` is the kernel_size,...
+1, same issue