About inference

[xiaoxing2001]

Hi~ Very good work, but I would like to ask the authors how to increase the number of frames processed during inference. It seems that currently only the first frame is processed. I encountered a dimensional mismatch error when trying to increase the number of video frames during inference.

Here, the modification I made in the .yaml file is: begin_clip_id: 0
end_clip_id: 1 --> begin_clip_id: 0
end_clip_id: 3

[Vicky0522]

Which item are you testing?

[xiaoxing2001]

item1

[Vicky0522]

You can try setting begin_clip_id=0, end_clip_id=2, and use_consistency_attention_control=True to extend the editing length.

[xiaoxing2001]

You can try setting begin_clip_id=0, end_clip_id=2, and use_consistency_attention_control=True to extend the editing length.

thank you, I'll try according to your advice

[xiaoxing2001]

I've encountered an another problem when I try to edit the second frame:

[xiaoxing2001]

Dear author, could you give a detailed explanation about the relationship between these parameters and the output video length?

[Vicky0522]

I've encountered an another problem when I try to edit the second frame:

Have you successfully extended the editing length? You can try the config_multi_chunks.yaml file I just uploaded under the item1 folder. It supports editing for two clips for item 1.

[Vicky0522]

Dear author, could you give a detailed explanation about the relationship between these parameters and the output video length?

start_t and end_t represent the start and end times (in seconds) for the editing window, while sample_fps indicates the number of frames sampled per second. In this case, we sample (8.4 - 0.0) * 2.5 = 21 frames from the source video.

chunk_size defines the number of frames in each clip for the SVD model, and overlay_size specifies the number of overlapping frames between clips. Here, we have two clips, each with a length of 11 frames, where the last frame of the first clip overlaps with the first frame of the second clip.

output_fps specifies the frames per second for the final edited output.

Finally, begin_clip_id and end_clip_id indicate the starting and ending clip indices. For example, if begin_clip_id=0, end_clip_id=2, the clip indices are in the range [0, 2). That means we edit the clips with indexes of 0 and 1.

[DanahYatim]

Hi, I'm not sure I understood this explanation, If I would like to edit a video with originally - 49 frames, 6 seconds, 8 fps.

What should I have for - start_t, end_t, sample_fps, chunk_size, output_fps, begin_clip_id, end_clip_id ?

Any other parameters that I should modify according to my input?

[Ethan-Li123]

Hi, I'm not sure I understood this explanation, If I would like to edit a video with originally - 49 frames, 6 seconds, 8 fps.

What should I have for - start_t, end_t, sample_fps, chunk_size, output_fps, begin_clip_id, end_clip_id ?

Any other parameters that I should modify according to my input?

Hi, have you solved it? What are your parameters?

[Vicky0522]

Hi, I'm not sure I understood this explanation, If I would like to edit a video with originally - 49 frames, 6 seconds, 8 fps.

What should I have for - start_t, end_t, sample_fps, chunk_size, output_fps, begin_clip_id, end_clip_id ?

Any other parameters that I should modify according to my input?

One of the plausible settings is as follows:
start_t = 0
end_t = 6
sample_fps = 8
chunk_size = 14
output_fps = 8
begin_clip_id = 0
end_clip_id = 3
overlay_size = 1

This will output an edited video with 14 + (14 - 1) * 2 = 40 frames in total.