Quesion about temporal part

[iseunghoon]

Thank you for the great resource. I have a question: It appears that, without considering the temporal axis, the features are fused independently for each frame in the form [bs*t, c, h, w]. Is that correct? There is no fusion of information between the frames except for video swin backbone. Did I understand right?

[bo-miao]

Yes, you are correct.