anhttran / 3dmm_cnn

Regressing Robust and Discriminative 3D Morphable Models with a very Deep Neural Network
Other
862 stars 254 forks source link

About ground truth 3d face shape #6

Closed qiexing closed 7 years ago

qiexing commented 7 years ago

It's very kind of you to share the model, I'am inspired by your work. I'am a new guy to 3d face construction and I want to reproduce your paper. I found your paper generate ground truth 3d face shape by paper: "Estimating 3D Shape and Texture Using Pixel Intensity, Edges, Specular Highlights, Texture Constraints and a Prior". Are you implement the method? Or where can i found this code to generate ground truth 3d face shape? Thanks!!

TalHassner commented 7 years ago

@qiexing Thanks for your interest in our work.

Though we were inspired by that paper, we actually used a simpler method to produce the training data labels, and we describe this method in our paper.

Basically, we estimate 3DMM parameters using our own implementation (which I don't think we'll be sharing, but I believe any standard 3DMM fitting method would do the trick here). We then pool these parameter vectors (element wise average) over multiple estimates from multiple images.

That's pretty much it. If you need a lot of 3DMM parameters for a lot of images, and don't want to implement their method, use CNN for this purpose J

I hope this helps.

qiexing commented 7 years ago

@TalHassner Your reply is really helpful. I want to more 3d parameter information, like expression parameter. Will you release your training data? If not, do you know some code which can produce 3d parameter by a given 2d images?

leeeeeeo commented 6 years ago

@qiexing Same question. Have you found any code to produce 3d parameter by given 2d images?

lyyiangang commented 3 years ago

same question. @leeeeeeo I find an repo named eos on github , it can generate 3dmm parameter from one image. not sure it's helpful for you.