Closed yyssmm closed 5 years ago
for (int i = 0; i < num_iterations; ++i) { vector<Eigen::Matrix<float, 3, 4>> affine_from_orthos;
vector<VectorXf> mean_plus_blendshapes;
for (int j = 0; j < num_images; ++j)
{
}
}
Do you write the enumerated pictures in the number of cycles in order to let 3D touch you to fit all the pictures?
But I have another idea, one by one(picture's landmarks), and the next with the 3D model parameters of the final result of the previous one. Are there different and better effects?
author please?
Hi!
@yyssmm, I would say that your landmarks are probably too inaccurate. I'd suggest you visualise the landmarks to double-check. Note that you will not get perfect alignment with landmark-fitting. I'd first check/improve the landmarks. Then, if they're good and you're still not happy, maybe you could use some post-processing (warping/optical flow)?
@sunjunlishi, both strategies would be valid and it would be interesting to compare them actually. If you have the images come in one-by-one, the second strategy might be better, while if you've got all the images in advance, optimisation over all of them might be beneficial.
thank you very much。it is good for reconstruct
@sunjunlishi how do you fiiting the 3d model through one by one image, can you share the code, i use fit_shape_and_pose, input last pca_shape_coefficients and last blendshape_coefficients.
yes。
@LeeTaiTai i have no trying
thank you for sharing your code. When I used the "fit-model-multi" project, I found the merged isomap had some flaws. For example, when I fit three images of a person to the model, the result isomap was wrongly alignment. Would you please help me to find out where is the problem. Thank you!