ik_vision

Hi, Thanks for your quick reply @luoyetx I run two training session in sequence. I run the beginning of the pbootstrap.sh code in order to train level 1: ``` python2.7...

Thanks for the [pdf](https://github.com/luoyetx/face-alignment-presentation/blob/master/Face-Alignment.pdf), I will try to reproduce its 1_F loss plot to understand why it fails. I will also try to validate the point's order in the augmented...

hi @yyytq I tried debugging the input loading and processing but didn't find its root cause or a solution for it

@wuqiangch I would recommend to pad the image by zeros (I use pad_size=10 pixels) to avoid boundary problems. `pad_img=numpy.pad(img, ((10,10),(10,10)),'constant')` After the point detection procedure, be aware to take off...

@layumi , from my understanding, in case the computer has no GPU matlab will create a warning with no side effects, as GpuArray will be converted to CPU on load....

@harvitronix I think that add multi-gpu support and NasNet can be a useful for the community, what needs to be done to merge those into the main fork?

I think the center at the origin is a feature of VIBE unrelated to the FBX output. The weak perspective assumption used in HMR/SPIN/VIBE, assumes that movement in the depth...

I think https://github.com/facebookresearch/eft can be helpful for occlusions, but it has a slow inference as it Exemplar Fine-Tuning in test time.  A short youtube video reviewing the method https://youtu.be/F2_SCM2Oqs4?t=1913

I think SPIN is a better fit for this application as it operates frame level. SPIN delay will be the processing time of the backbone of a single image, not...

Most of the compute time per frame is spent in the backbone (resnet50 in HMR), therefore removing the mesh generation (in SMPL model) is not likely to get a significant...