Training PoseNet in Python #1389
Description
To get help from the community, we encourage using Stack Overflow and the tensorflow.js tag.
TensorFlow.js version
Posenet for python https://github.com/rwightman/posenet-python
Browser version
Describe the problem or feature request
I had been using the Pose-net implementation, instead of the js the python one, but I would like to use my own training model instead of the one given by google, as I think that in low light and noisy images, changing the model to a model develop with bodies labeled in the same condition can improve significantly the detection. I am currently using this implementation https://github.com/rwightman/posenet-pytorch, with minimal changes to detect only face and arms. I had found that it is possible to convert from caffe model to tensorflow but nothing about using own images to train Pose-net, or using other models in pose-net apart from the one given by google. This is because I want to use it to detect the joints in low light and with realtively cheap cameras which has a bit noisy images.
I tried pre-procesing the images with brightness, gain and contrast changes but the result is nearly the same, reason from which I though of changing the model to a one labeled with this type of images.