the video demos in https://orient-anything.github.io/ are smooth

[qinghew]

may I ask what smoothing operation is used? could you share this part of the code for videos?

[zhang-ziang]

Thank you for your interest in this project!
We extract the video frame by frame and feed each image into the model.
Yes, we post-processed the output using a simple averaging operation in context, which was able to reduce the slight jitter of the rendered axis.
We'll release the pipeline for videos later.
:)

[qqzy]

Thank you for your interest in this project! We extract the video frame by frame and feed each image into the model. Yes, we post-processed the output using a simple averaging operation in context, which was able to reduce the slight jitter of the rendered axis. We'll release the pipeline for videos later. :)

Hi @zhang-ziang
Thank you for the explanation about the post-processing technique used in the video demos. I'm interested in implementing this in my own projects. Could you share more details or a rough timeline regarding when the post-processing code will be available