Hey, I've checked the demos you implemented, impressive work! Thanks for sharing.

I'd like to ask, where did you get the weights in blazepoze full body ?

I have checked this repo but it appears to me that there are only weights for the upper_body model and the pose detector over there.

Thanks in advance!

In

This should be: 'eyeRight', 'eyeLeft', 'nose', 'mouthRight', 'mouthLeft'

i added an option to use png face mask with transparency for the faceswap demo. this is useful when the mask contour exceeds the actual face and includes some of the background. in this case, allowing a transparent background in the mask allows a much cleaner effect. i changed in https://github.com/eyaler/tfjs_webgl_app/blob/master/face_landmark/render_facemesh.js#L32

void main(void)
{
    vec4 color;
    color = vec4(texture2D(u_sampler, v_texcoord));
    color[0] *= u_color[0];
    color[1] *= u_color[1];
    color[2] *= u_color[2];
    color[3] *= v_vtxalpha * u_alpha;
    gl_FragColor = color;
}

you can see it in work in my fork for offline videos: https://eyaler.github.io/tfjs_webgl_app/face_landmark
thanks for your awesome work! i learnt a lot from you!

I have reported an error after I run your project. I look up this question online, saying that my element is not set to high. But I have seen the DOM, how to solve this problem.This is an error message
Requested texture size [0x0] is invalid

Hey!

First of all, thank you very much for sharing your work, it has helped guide my project more than you can know.

I have a question about full body joint estimation.

If I'm not wrong, the landmark model provides (x,y,z) triplets as outputs. However you only seem to be using the (x,y) coordinates in post processing and drawing of joint key points. Why is that? Shouldn't one be computing a 2D projection using the z dimension as well?

Thanks in advance!

I click on the home page of handpose and display failed to load model after loading for a period of time