search

jeeliz / jeelizFaceFilter

Javascript/WebGL lightweight face tracking library designed for augmented reality webcam filters. Features : multiple faces detection, rotation, mouth opening. Various integration examples are provided (Three.js, Babylon.js, FaceSwap, Canvas2D, CSS3D...).
https://jeeliz.com
Apache License 2.0
2.68k stars 533 forks source link

Issues in including external model(except that from listed in NN) to JeelizFaceFilterAPI #207

Closed nandu19981998 closed 3 years ago

nandu19981998 commented 3 years ago

Hi,

I'm currently working with a web based application that enables virtual TryON, mainly composed of superimposing 3D objects to earpoints in face. I was successful in using NN_default.json model to extract the neckpoints, but I like to have earpoints as well.

Is there any way in which I could include the model that I have developed that identifies both earpoints and neckpoints to the JeelizFaceFilterAPI.

Thanks in advance!!!

xavierjs commented 3 years ago

Hi,

We use a proprietary file format for the neural network model since I use my own toolchain to create and train neural networks (I don't use TF or Torch or Keras). Moreover, this library does not extract the face landmarks positions before computing the pose; but directly get the pose from neural network output. The pose is computed to be aligned with a virtual point between the 2 eyes and scaled according to the distance between the 2 ears. The neck is near the pose central point so you can compute the neckpoints as you do with current output data quite accurately. but ears are too far to get accurate positioning.

I have another library called WebAR.rocks.face which outputs face landmark position. Different NN models are provided and demos, including earrings virtual try-on. I charge for it ( less than my competitors and there is no cost per user session ). You can contact me at contactatwebar.rocks if you are interested.

nandu19981998 commented 3 years ago

Thanks for your reply, will look into it and get back to you

On Tue, 5 Jan, 2021, 9:41 pm Bourry Xavier, notifications@github.com wrote:

Hi,

We use a proprietary file format for the neural network model since I use my own toolchain to create and train neural networks (I don't use TF or Torch or Keras). Moreover, this library does not extract the face landmarks positions before computing the pose; but directly get the pose from neural network output. The pose is computed to be aligned with a virtual point between the 2 eyes and scaled according to the distance between the 2 ears. The neck is near the pose central point so you can compute the neckpoints as you do with current output data quite accurately. but ears are too far to get accurate positioning.

I have another library called WebAR.rocks.face which outputs face landmark position. Different NN models are provided and demos, including earrings virtual try-on. I charge for it ( less than my competitors and there is no cost per user session ). You can contact me at contactatwebar.rocks if you are interested.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/jeeliz/jeelizFaceFilter/issues/207#issuecomment-754733797, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALNF7ADONNW4NX5ZVEMMSSTSYM2ZTANCNFSM4VOIZL2A .