Problem producing high resolution output

[oztrkoguz]

In order to get high resolution output in the project, it is necessary to use use_sr=True, but there should be srmodel in it, but this file is not uploaded, will you share it?

[nicholaspei]

You should run this demo , I've updated the code, and now I can get high resolution output video,my example video is :1080 × 1920
https://github.com/nicholaspei/video-retalking/blob/main/quick_demo.ipynb

[nicholaspei]

@ooguzzzz

[oztrkoguz]

Thank you for your interest, I tried what you said, I got results by giving input in 1920x1080 size, there was no improvement in the results as I wanted, the outputs displayed in the project give much better results, how can I do this?
@nicholaspei

[conkeur]

@ooguzzzz

there are 2 laplacian pyramid in the whole process ,one is the afore mentioned by @nicholaspei . the other in the enhance.process() in inference.py row 267. you can change both resize before these pyramid process from 512,512 to 1024,1024 or 2048,2048. and open possion blending in the enhancer.process() .image should look better with cost of more time

[avinash403]

@ooguzzzz

Since your code calls GPEN-BFR-1024 model, where do you get it? The google drive mentioned in the repo doesn't have it

[avinash403]

@ooguzzzz

Since your code calls GPEN-BFR-1024 model, where do you get it? The google drive mentioned in the repo doesn't have it

Nervermind! if someone else is looking for the same model, it can be found here: https://www.kaggle.com/code/thangclassic/retalking

[jakeytan]

original:

aaa_readme_00024.mp4

general results:

output_346.mp4

[hidigiasia]

@jakeytan Nice! Is this the result of video-retalking? or you make some changes to the code?

[ywythu]

original:

aaa_readme_00024.mp4
general results:

output_346.mp4

do you make some changes? Since my output breaks and changes to the original pose every 5 minutes, I dont know how to fix that