search

MRzzm / DINet

The source code of "DINet: deformation inpainting network for realistic face visually dubbing on high resolution video."
895 stars 167 forks source link

My own data, the effect is very poor, come with a mask, this is why #77

Open zhanghongyong123456 opened 6 months ago

zhanghongyong123456 commented 6 months ago

result： image src： image

Inferencer commented 6 months ago

result： image src： image

Lighting & skin color are the reasons, even if you fixed both you would still have an issue with the facial hair

zhanghongyong123456 commented 6 months ago

result： image src： image

Lighting & skin color are the reasons, even if you fixed both you would still have an issue with the facial hair

Is there any other good driving method

Inferencer commented 6 months ago

answered in the optimized repo

JoraAlgorithm commented 6 months ago

@Inferencer optimized repo? please send us link

Inferencer commented 6 months ago

@Inferencer optimized repo? please send us link

I don't use it as I prefer the current state of DINet & it's results using DeepSpeech although one of the main benefits of the optimized repo is the removal of the openface steps & the plans to incorporate the latest Deepspeech model in a few weeks/months https://github.com/Elsaam2y/DINet_optimized

JoraAlgorithm commented 6 months ago

Sure, Thank you very much. If you require a hand, please do inform me.