Spatially Enhanced Cross-Attention

[yunfeiheng]

Recently, I have been studying this paper and I am not very clear about formula 4 in the paper. The spatial weight map M is generated for each type of manipulation operation. Therefore, if there are several annotations in the manipulation sequence, there should be several M. So, in formula 4, M should be the sum of all M for each operation of each input? I am not very clear about this calculation.

[mamarasulovs]

Recently, I have been studying this paper and I am not very clear about formula 4 in the paper. The spatial weight map M is generated for each type of manipulation operation. Therefore, if there are several annotations in the manipulation sequence, there should be several M. So, in formula 4, M should be the sum of all M for each operation of each input? I am not very clear about this calculation.

write down by wechat :sms9299

[islimeng]

i have a configuration quesion but i cannot reslove it,would someone can help me?

Traceback (most recent call last):
File "D:/seqdeepfake/SeqDeepFake/train.py", line 435, in
cfg = Config(args.cfg)
File "D:\seqdeepfake\SeqDeepFake\models\configuration.py", line 6, in init
with open(config_file) as f:
TypeError: expected str, bytes or os.PathLike object, not NoneType