fq

a litte problem in fpn code

3

in line 45: p3 = F.interpolate(p4, size=[p4.size(2), p4.size(3)]) + p4 should change to: p4 = F.interpolate(p5, size=[p4.size(2), p4.size(3)]) + p4

could you plan to update the trainning code ? FCOSTargetGenerator is not there, When will you do it well?

will open the training code?

7

统一拓扑结构得到了约2000个不同性别、年龄、种族的人脸mesh，在此基础上，构建了一个全头人脸3DMM并命名为HIFI3D++. all these data preprocess script will be release?

can replace the render code to pytorch3d?

1

opendr is ... its not a problem, just ... as you konw.

about data pre process

1

您好，我测试了一些自己的模型，增加了normalize=1，发现骨骼尺度上有差距，最终出来的蒙皮权重就很差，有几个问题想问下： 1. 三角化后的模型consistent upright and front facing orientation.是指模型导入引擎，朝向(比如fbx脚本里的axis_forward='-Z', axis_up='Y')一致，然后是T-pose就行？ 还是三角面片坐标上有什么含义？ 2. 和smpl_std.obj对齐，直接使用您给定的归一化函数就可以？还是有什么更细致的要求？ 

like 

the code.

is there some resource? or some fake code? https://www.mathworks.com/help/vision/ref/estimategeometrictransform3d.html https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.transform.Rotation.align_vectors.html http://learning.aols.org/aols/3D_Affine_Coordinate_Transformations.pdf https://en.wikipedia.org/wiki/Kabsch_algorithm

one new project of lsp for complete training and deploy can look at here?

1

consider the security/licence of original project,i recreat the alg base on deca and LiveSpeechPotraits and the other block is comming. there are still some problem, welcome discuss and pr. here:...

in inference, your camer model is fixed, which means camera_intrinsic, sacle etc is constant. but when i use deca, every image has its own camera paremeters, so in traing data,...