Does this method requires detect hands get centers first when prediction?

Question

Does this method requires detect hands get centers first when prediction?

Closed this issue 5 years ago · 4 comments

lucasjinreal commented 5 years ago

Shall we need detection centers when prediction?

Answer 1 · 2019-05-30T08:53:27.000Z

The output of the model is a normalized representation, so you need a center to transforn it to the world coordinate.

Answer 2 · 2019-05-30T11:53:15.000Z

@dumyy What do you mean need a center to transform to world? Does it need a center as network input?

I just wonder the network can using a single image input and output can be draw on image?

Answer 3 · 2019-05-30T12:27:54.000Z

the input of network is only a single depth image, output is normalized 3d coordinates. So you need transform them into real world coordinates by post-processing. If you want to draw the output on image, just map the 3d pose to 2d image by camera info.
ALL THOSE CAN BE FOUND IN THE TEST FILE.

Answer 4 · 2019-05-30T12:30:23.000Z

thank u so much.... so this method still need depth.... I am searching for a method simply using RGB image to do so.