Bad Results.

[maocaixia]

How to improve it. Any Suggestion.

The more similar the video and reference picture are, the better the effect will be.

https://github.com/magic-research/magic-animate/assets/23650992/c582ba89-9b38-489c-8346-70d4d79bf5ca

https://github.com/magic-research/magic-animate/assets/23650992/50ccf757-98e5-48c7-a4d2-75a9ce568b27

https://github.com/magic-research/magic-animate/assets/23650992/3ee17c26-11c0-4918-8da8-9cf8b2dde287

[ValentinaILove]

Hola soy nueva por aqui....saludos a tod@s , como puedo ejecutar el programa bajo Windows 11 para hacer unas pruebas

Gracias

[FurkanGozukara]

same here can't get good results either

[Echolink50]

Last one looks pretty good. Not sure what people are expecting for a first release.

[FurkanGozukara]

Last one looks pretty good. Not sure what people are expecting for a first release.

i think it gives good results only with what was their training dataset

have you got any good results with custom densepose?

[luccachiang]

Last one looks pretty good. Not sure what people are expecting for a first release.

i think it gives good results only with what was their training dataset

have you got any good results with custom densepose?

Hello, thanks for your contribution to this project! Have you ever tried this model on your custom dataset? Since the paper gave some generalization results, it looks fine with unseen data. Btw, if the identity is changed, say from human to animals, does the model need to be retrained? It seems that the author did not teach us how to train from scratch.

[FurkanGozukara]

Last one looks pretty good. Not sure what people are expecting for a first release.

i think it gives good results only with what was their training dataset have you got any good results with custom densepose?

Hello, thanks for your contribution to this project! Have you ever tried this model on your custom dataset? Since the paper gave some generalization results, it looks fine with unseen data. Btw, if the identity is changed, say from human to animals, does the model need to be retrained? It seems that the author did not teach us how to train from scratch.

i never tried training

[dedkamaroz]

I must say I'm pretty disappointed in the results so far when using my own generated densepose clips. I made sure to have both face photo and densepose animation clip at 512x512 resolution and tried all combos of 5 different faces x 3 densepose clips, and all of them look at best cartoonish, and most suffer from an obviously confused AI's interpretation of the animation..

Of course I do appreciate the efforts of the developer and overall it's an amazing piece of software, but I now get why the example gifs are so tiny :)

[FurkanGozukara]

I must say I'm pretty disappointed in the results so far when using my own generated densepose clips. I made sure to have both face photo and densepose animation clip at 512x512 resolution and tried all combos of 5 different faces x 3 densepose clips, and all of them look at best cartoonish, and most suffer from an obviously confused AI's interpretation of the animation..

Of course I do appreciate the efforts of the developer and overall it's an amazing piece of software, but I now get why the example gifs are so tiny :)

:D