How to implement rope encoding of the reference image

[sunjian2015]

The paper mentioned that there are changes in the rope encoding of the reference image, where is the code implementation?

[sunjian2015]

I checked the code, here it seems that only the offset to w is achieved, no h, means RoPE_z(f, i, j) = RoPE(-1, i + w, j), not the same as mentioned in the paper, why?

HunyuanCustom/hymm_sp/helpers.py

Line 71 in abffb8f

elif concat_dict['mode']=='timecat-w':