confusion in implementation of decoded network

[mrluin]

Hi,
Thanks for your implementation!
I have some confusion in the implementation of the decoded network.
I knew that each cell has two inputs from prev_prev_cell and prev_cell. But what the case in the decoded network. I mean that: If layer_i has downsampling rate 16, but layer_{i-1} has downsampling rate 4. In this case, prev_prev_c is None and only has prev_c's output as input, right? (from the paper, I saw that prev_prev_c's output should have the same scale as the current layer).
Could you give me some hints? Thanks in advance.

[zhizhangxian]

If lay_i havs downsampling 16 then it will only get information from lay_{i-1} with sample 8 and 16 not 4

[mrluin]

Thank you for your quick answer! @zhizhangxian
I have another question about affine of BN in train_search phase and retrain phase. From darts, the learnable parameters of BN are disabled in train_search phase. But in some other paper like proxylessNAS, affine of BN is not set specifically. What effect do you think affine mode of BN might have?
Looking forward to your reply!

[zhizhangxian]

In our experiment, we set affine to False in search and True in retrain
False in search is accroding toi darts
True in retrain is accroding to deeplab v3+