Where is the Flex Projection Layer？

[Harryx2019]

After carefully reviewing the code, I found that the Flex Projection Layer mentioned in the paper does not appear in the implementation, for example:

• In the GitHub repository’s zero_shot.py, PatchCB directly specifies patch_len.
• In the HuggingFace repository’s modeling_LightGTS.py, the input z is resized, which is essentially a linear interpolation operation.

So, where exactly is the Flex Projection Layer that the paper refers to implemented?