Question: concat_image and ops

[yjhong89]

Hi!

1. First question is what purpose of "concat_image" ?

• CogVideo/sat/dit_video_concat.py

  Line 823 in bbe909d

  if "concat_images" in kwargs and kwargs["concat_images"] is not None:

• I thought it would be used for image condition for I2V, but image condition is fed as "concat" condition in here. (actually "concat_image" is not used in sat/sample_video.py)

  • CogVideo/sat/sample_video.py

    Line 209 in bbe909d

    c["concat"] = image

• So I wonder what is the purpose of "concat_image"

2. What is "ofs_emb" for?

• CogVideo/sat/dit_video_concat.py

  Line 840 in bbe909d

  if self.ofs_embed_dim is not None:

• It is added to timestep embedding and it is set to 2.0 at image2video.

  • CogVideo/sat/sample_video.py

    Line 214 in bbe909d

    samples_z = sample_func(